Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

Subscribe now

POLITICS

• The Trump administration ordered federal agencies to stop using AI systems developed by Anthropic, escalating an ongoing dispute over military use of the company’s models.

  • The order includes penalties following a phased wind-down period for agencies currently relying on Anthropic’s technology, including the Department of Defense.

  

  • The Department of Defense formally designated Anthropic a “supply chain risk,” a classification typically reserved for foreign adversaries.

    • Anthropic has said it plans to challenge the designation.

• Shortly after the Anthropic decision, OpenAI announced a new agreement with the Pentagon to supply AI systems for classified government networks.

  • The deal reportedly includes restrictions on mass domestic surveillance and fully autonomous weapons systems.

    • Over 700 current employees from OpenAI and Google signed a petition opposing Pentagon efforts to pressure AI firms into expanding military use of their models.

    • The petition cites concerns over surveillance, autonomous weapons, and erosion of company safety commitments.

• The Trump administration repealed Biden-era Mercury and Air Toxics Standard.

  • The EPA has said that this move aims to alleviate utility costs as data center power demand grows.

• NIST and CAISI announced a new AI agents standards initiative.

• Reps. Brian Babin (R-TX) and Jay Obernolte (R-CA) have requested that the Government Accountability Office produce a report on the current regulations concerning AI policy at the state and federal level.

  • This request comes in response to ongoing uncertainty and attempts by the White House to preempt or disrupt state AI legislation.

• Office of Science and Technology Policy Director Michael Kratsios announced a new “Tech Corps” initiative aimed at delivering “the American AI stack to the world”.

  • The new initiative was announced at the India Impact AI Summit on Feb. 20.

  • Kratsios represented the US at the summit, echoing sentiments made by J.D. Vance at last year’s summit in Paris that AI policy discussions should shift emphasis “from safety to opportunity.”

• A pro-AI advocacy group aligned with Republican political fundraising efforts claimed it has signed up more than 500,000 supporters nationwide.

• The Trump administration has instructed U.S. diplomats to push back against foreign data sovereignty and localisation laws that could restrict how American tech companies handle data abroad.

• Sens. Maria Cantwell (D-WA) and Todd Young (R-IN) reintroduced a bipartisan AI innovation bill aimed at strengthening U.S. leadership in artificial intelligence and related technologies.

• The White House has engaged with Florida lawmakers to oppose the state’s Artificial Intelligence Bill of Rights, a proposal that would impose new limits on AI use and strengthen protections for minors.

Subscribe now

TECH

• OpenAI announced a $110 billion funding round — the largest private tech financing in history — valuing the company at $730 billion pre-money.

  • Amazon led with $50 billion, with Nvidia and SoftBank contributing $30 billion each.

• Google’s newly released Gemini 3.1 Pro is boasting record benchmark scores.

• Google also added music-generation to the Gemini app, using DeepMind’s Lyria 3 model to deliver the new capability.

• OpenAI disclosed that they had flagged the Tumbler Ridge shooter’s chats months prior to the mass shooting.

  • The company banned Van Rootselaar’s account but did not find that the activity met the “higher threshold required” to report it to law enforcement.

  • OpenAI has since published a report on the detection and prevention of malicious uses of AI.

• In the ongoing Musk v. Altman case, OpenAI filed a motion to exclude expert testimony from AI safety professor Stuart Russell, labeling him a "prominent AI doomer" whose views are "dystopian" and "alarmist."

• OpenAI announced "Frontier Alliances" — multi-year partnerships with consulting firms Accenture, BCG, Capgemini, and McKinsey — to help enterprise customers adopt its Frontier AI agent platform.

• Anthropic accused Chinese labs DeepSeek, Moonshot AI, and MiniMax of running "industrial-scale campaigns" to extract Claude's capabilities via over 16 million exchanges generated through approximately 24,000 fraudulent accounts.

• OpenAI announced that ChatGPT has reached 900 million weekly active users, alongside 50 million paying subscribers.

DISCOURSE

• Is AI already killing people by accident? (Gary Marcus) ✍

• Sorry skeptics, AI really is changing the programming profession (Understanding AI) ✍

• The Pentagon is making a mistake by threatening Anthropic (Understanding AI) ✍

• The least understood driver of AI progress (Epoch AI) ✍

• How worried should we be about AI biorisk (Transformer) ✍

• I’m not convinced the “A.I. backlash” is real (Matthew Yglesias) ✍

• How much does distillation really matter for Chinese LLMs? (Interconnects AI) ✍

• Full interview: Anthropic CEO responds to Trump order, Pentagon clash 📽

• The Moving and the Still - Reflections on Delhi (Dean Ball - Hyperdimensional) ✍

• The case for wasteful agents (The AI Frontier) ✍

• The Race to Build God: AI’s Existential Gamble (Your Undivided Attention) 🔉

• Why Teaching AI Right from Wrong Could Get Everybody Killed - Max Harms (80,000 Hours Podcast) 🔉

• The AI Skeptics (The Last Invention) 🔉

• Why we need a moratorium on superintelligence research (Transformer) ✍

• ✨⚕️ Must AI cure cancer? (Faster, Please!) ✍

• 🔮 Where the human ends and AI begins (Exponential View) ✍🔉📽

Subscribe now

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

Subscribe now

POLITICS

• POLITICO reported that the Trump administration has drafted a pact with the aim of ensuring data center build outs do not increase household energy prices.

  • Anthropic have since committed to covering any consumer energy price increases resulting from its data center expansion:

  

  Source: Anthropic

  • Sens. Josh Hawley (R-MO) and Richard Blumenthal (D-CT) introduced legislation with the same goal. The GRID Act would ban new grid-connected data centers and mandate that data centers publish current and future usage.

• The White House pressured Republican Utah Senate Majority Leader Kirk Cullimore Jr. to abandon AI transparency legislation.

• Rep. Josh Gottheimer (D-NJ) introduced the bipartisan AI Workforce Training Act, which would create a federal tax credit for companies that provide AI training to their employees.

• Sens. Elizabeth Warren (D-MA) and Jim Banks (R-IN) are reportedly preparing bipartisan legislation banning the export of certain chips to China, in defiance of President Trump’s increasing leniency.

Subscribe now

TECH

• A mixed week for Anthropic: the company lost a key figure while simultaneously securing a massive new funding round.

  • A notable departure as safeguards research lead Mrinank Sharma resigned. In a letter posted on X, he wrote

    “I’ve repeatedly seen how hard it is to truly let our values govern our actions.”

  • The resignation comes as Anthropic released its latest Sabotage Risk Report for Opus 4.6, which found the model shows an “elevated susceptibility” to misuse for “heinous crimes.”

  • At the same time, Anthropic closed its Series G round, raising $30 billion at a $380 billion valuation — a sharp increase from its previous $183 billion valuation.

  • The company also pledged $20 million to AI safeguards group Public First Action, led by former Representatives Chris Stewart (R-UT) and Brad Carson (D-OK).

  • Additionally, the Wall Street Journal reported that the Pentagon used Anthropic’s Claude to aid the capture of Nicolás Maduro, via their partnership with Palantir.

    • The Pentagon has this week threatened to cut ties with the tech firm following months of failed negotiations to secure an agreement on using Claude for “all lawful purposes.”

• Departures continue at xAI, as co-founders Tony Wu and Jimmy Ba announced they were leaving the company. They are the fourth and fifth founders to leave in the last year.

• This came as the company’s CEO, Elon Musk, announced a restructuring that includes a strategic pivot from Mars exploration toward a renewed focus on the Moon.

• OpenAI have also seen departures as researcher Zoë Hitzig resigned over the introduction of ads. You can read her NYT op-ed here.

• The company also permanently retired its controversial 4o model, along with several other models.

• Apple’s upgrade to Siri, powered by Google Gemini, has been pushed back.

• Meta aims to add 1GW of capacity with its new data center in Lebanon, IN.

DISCOURSE

• Do Not Surrender to the Tech Tree (Tao Burga, Macroscience) ✍

• What do “economic value” benchmarks tell us? (Epoch AI) ✍

• Dario Amodei — “We are near the end of the exponential” (Dwarkesh Podcast) 📽

• AI Doesn’t Reduce Work—It Intensifies It (Harvard Business Review) ✍

• Why the AI industry can’t resist dirty on-site gas turbines (Transformer) ✍

• On Recursive Self-Improvement (Part II) (Hyperdimensional) ✍

• Why China Is Falling Behind in AI | by PKU Scholar Hou Hong (Sinification) ✍

• CPUs are Back: The Datacenter CPU Landscape in 2026 (Semianalysis) 📊

• What’s Real And What’s Fake In Tech (Core Memory) 📽

• The singularity won’t be gentle (Nate Silver) ✍

• Steve Yegge on AI Agents and the Future of Software Engineering (The Pragmatic Engineer) ✍

• The Scientist and the Simulator (Latent Space) ✍

• Don’t let Elon Musk monopolize space compute (Slow Boring) ✍

• The best international AI policy is to fix national strategies (Anton Leicht) ✍

• Magnitudes of intelligence (Exponential View)

• The Origins and Limitations of AMD’s Historic Revival (The Chip Letter) ✍

• The $200 Billion Bet (Chipstrat) ✍

• Seedance, Kling and the Chinese AI Video Ecosystem (ChinaTalk) ✍

Subscribe now

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

Subscribe now

POLITICS

• President Trump announced Project Vault, “a critical mineral stockpile for American businesses.”

• Civil society and policy groups wrote a letter to Senate leaders opposing AI preemption.

• Sens. Cotton (R-AR) and Klobuchar (D-MN) introduced The Biosecurity Modernization and Innovation Act, requiring “gene synthesis providers screen their orders and customers for bad actors or dangerous pathogens.”

  • Americans for Responsible Innovation endorsed the legislation, citing AI’s role in accelerating biosecurity risks.

• A coalition of non-profits urged a Grok ban in federal agencies following the controversy surrounding its generation of harmful sexualized images.

• New York is considering a pair of bills to restrict AI industry and usage.

  • The first bill imposes a three-year moratorium on new data center construction.

  • The second (“NY FAIR News Act”) would require sites to label AI generated news, and mandates that any news content created by AI must be approved by a human with “editorial control.”

• A report by the Center for Data Innovation and Public First found that the U.S. is an “uneven adopter” of AI use in government.

  • The report surveyed thousands of government workers across ten countries.

TECH

• Anthropic released Opus 4.6 with new “agent teams” feature. The new addition uses teams of agents to break down larger coding tasks into more manageable subtasks.

  • OpenAI also released their newest coding model GPT-5.3-Codex, which they claim is their “first model that was instrumental in creating itself.”

• OpenAI is also reported to be unhappy with Nvidia, and looking for alternative suppliers for its chips.

  • At the same time, Nvidia’s $100 billion investment in OpenAI, announced in September, seems to be off the table.

  • Sam Altman responded on X that they “love working with Nvidia” and “hope to be a gigantic customer for a very long time.”

• Gemini surpassed 750 million monthly users ahead of Gemini 3 launch.

• Intel announced it will begin producing GPUs.

• Meta announced in an emailed statement that it would be testing a standalone Vibes app.

• Palantir CTO Shyam Sankar says that “the American people are being lied to about AI”, arguing that it should be seen as a tool that empowers workers, not replaces them.

• SemiAnalysis discussed the ‘once-in-four-decades’ shortage fueling a ‘memory boom.’

• Nathan Lambert compared Opus 4.6, Codex 5.3, and discussed the difficulty of evaluating models in 2026.

• International AI Safety Report was released, led by Yoshua Bengio.

Subscribe now

DISCOURSE

• Data to start your week (Exponential View) 📊

• Brain-Inspired AI Chips | $4.5B Unconventional AI (Sourcery)

• How close is AI to taking my job? (Epoch AI) ✍

• Do AI models cover their costs? (Epoch AI x Exponential View) 📽 🔉

• 10 articles that set the tone for AI policy in 2025 (AI Policy Bulletin Newsletter) ✍

• An Allied World on the American AI Stack: A Strategy for Export Leadership (Anton Leicht, Sarosh Nagar, Liam Patell and Sam Winter-Levy) ✍

• How to build a data moat (The AI Fronter) ✍

• On Recursive Self-Improvement (Part I) (Dean Ball) ✍

• “Build the AI Wall” with Steve Bannon (The Last Invention) 🔉

• AI manipulation (AI Policy Perspectives) ✍

• In both China and the US, compute makes up more than 50% of the cost of running an AI company (Epoch AI) ✍

• Elon’s entire tech tree is converging right now (Dwarkesh Patel) 📽

• Experts Have World Models. LLMs Have Word Models. (Latent Space) ✍

• LLM societies; Huawei makes kernels with AI; ChipBench (Import AI) ✍

• The many masks LLMs wear (Understanding AI) ✍

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

POLITICS

• Sen. Markey (D-MA) questioned OpenAI, Anthropic and others about the use of advertising in chatbots, describing it as a “significant, and potentially dangerous transformation.”

• Rep. Obernolte (R-CA) has a “forthcoming” billed being referred to as the “Great American AI Act.”

• Director of OSTP Kratsios called for Congress to back Trump’s AI strategy.

• Department of War launched the AI Acceleration Strategy, to “extend our lead in military AI deployment and establish the US as the world’s undisputed AI-enabled fighting force.

  • US Air Force is partnering with Stanford to test AI copilots.

  • Congress is reviewing agentic AI capabilities for use in defense.

• Former FCC Chair Wheeler discussed three ‘chokepoints’ that will decide the AI era.

  This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

  ---

TECH

• The Three A’s of Open AI:

  • Ads: OpenAI is officially bringing ads to ChatGPT.

  • Arguing: Their feud with Elon Musk continued, in a blog post on “the truth Elon left out.”

  • Answering: Questions from Congress.

  • Also:

    • Enterprise customers now account for 40% of the company’s business.

    • GPT-5.2 Pro set a new record on FrontierMath.

      

• At Davos, Anthropic CEO Dario Amodei took shots at Trump’s policy on Chips and China.

• Thinking Machines, former OpenAI CTO Mira Murati’s AI startup, is in meltdown.

• Elon announced that the Colossus 2 supercomputer is now live.

• Two Chinese AI companies, Zhipu and MiniMax, went public.

• The latest trends in AI, courtesy of EpochAI:

  • LLM inference prices are falling at 40x per year.

  • Total computing power of NVIDIA’s chip stock is growing at 2.3x per year.

  • OpenAI’s revenue is growing 3.2x per year.

  • The median time to build a gigawatt data center is 2.1 years.

• Nathan Witkin argues that the METR’s “time-horizon of AI tasks” benchmark isn’t all it’s cracked up to be.

• H100 prices hit an 8-month high.

• Zipline raised $600M+ at a $7.6B valuation.

Subscribe now

TOP READS

• Beware the Authoritarian Biopolitics of the AI Age (Sinification) ✍️

• Winning the era of superhumanity (Scott Belsky, Implications) ✍️

• Event: America’s Tech Long Game for Competing with China (CSIS) 🗓️

• How AI Safety Is Getting Middle Powers Wrong (Anton Leicht) ✍️

• Silicon Valley must decide which side it’s on (Brian Merchant) ✍️

• Tyler Cowen interviews Francis Fukuyama (The UnPopulist) 🎧

• Building Self-Aware AI Would Be a Bad Idea (AI Policy Bulletin) ✍️

• Jon Hartley and Arthur Laffer on California’s new wealth tax (WSJ) 📊

• Behavioral biases of AI models (Bini et al.) 📊

• How AI Reaches the Most Extreme Parts of the World: From U.S. Navy Ships & Saudi Deserts to the Alaskan Arctic (Sourcery) 🎥

• Winners and losers in the AI economy (Jack Clark, Anthropic) ✍️

• Intel’s long series of unforced errors (The Chip Letter) ✍️

• AI’s Harm By The Numbers (Harry Law) 📊

• How shifting risk to users makes Claude Code more powerful (Understanding AI) ✍️

• AI in 2026 (Tomas Pueyo) ✍️

• On AI and children (Dean Ball) ✍️

• How Reflect Orbital Is Building Satellites to Redirect Light From Space (The Generalist) 🎥

• Anthropic’s Head of Economics on AI adoption, Claude Code, the burden of knowledge & future experts (Azeem Azhar) 🎥

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

POLITICS

• A judge has ruled that OpenAI must turn over 20 million ChatGPT logs:

  • News outlets including the New York Times and Chicago Tribune had requested the logs as part of their ongoing copyright dispute with OpenAI.

  • The model developer had appealed the request, but Magistrate Judge Ona T. Wang ruled against them.

• The FDA announced that it will relax regulations on digital health products, allowing wearables and generative AI tools into clinical workflows.

• State Sen. Steve Padilla (D-CA) introduced a bill that would place a four-year ban on the sale and manufacture of toys with AI chatbot capabilities for kids under 18.

• The Consumer Federation of America (CFA) issued an “urgent call” for an investigation of Elon Musk’s xAI over the creation of non-consensual sexual abuse material.

  ---

TECH

• SoftBank completed its $40B investment in OpenAI

  • OpenAI’s deal with the Japanese firm places it at a $260B pre-money valuation.

• Meta’s chief AI scientist Yann LeCun left Meta with some parting shots on his way out:

  • LeCun said Meta’s newest AI boss, 29-year-old billionaire Alexandr Wang, was “young” and “inexperienced.”

    • Wang has “no experience with research or how you practice research, how you do it. Or what would be attractive or repulsive to a researcher,” he said.

• Claude Code said that 100% of his coding contributions to the model last month were written by… Claude Code.

• Up to 1/3 of YouTube’s feed may be AI slop.

• WSJ let Claude run their office vending machine - it lost hundreds of dollars.

• OpenAI co-founder Andrej Karpathy: “I’ve never felt this much behind as a programmer.”

• Microsoft CEO Satya Nadella shared his views and predictions for the upcoming year.

• “xAI has bought a third building called MACROHARDRR. Will take @xAI training compute to almost 2GW.”

• The 2025 leaderboard for AI models is in.

Subscribe now

TOP READS

• Claude Code is about so much more than coding (Transformer News)

• Big Tech v. the Chinese Government in New AI Companion Regs (ChinAI Newsletter)

• AI 2026: Data Centers Restart Growth of a Stagnant U.S. Electrical Grid (Bismarck Brief)

• How AI Labs Are Solving the Power Crisis: The Onsite Gas Deep Dive (SemiAnalysis)

• On Nvidia’s acquisition of AI chip startup Groq (The Chip Letter)

• The Race between Waymo, Cybercab, and Uber (Uncharted Territories)

• 9 AI predictions for 2026 (Transformer News)

• 17 more predictions for AI in 2026 (Understanding AI)

• Self-driving cars aren’t nearly a solved problem (Strange Cosmos)

• The AI copyright question has no easy answers (Transformer News)

• The AI Competition as a Football Game (Interconnected)

• Governments are struggling. Can AI help? (AI Policy Perspectives)

• Data to start your week (African solar, space traffic, and token economics) (Exponential View)

• A skeptical view of AI for biology (Bo Wang)

• Notes on Taiwan (Sam Enright)

• The Compute Theory of Everything (Samuel Albanie)

• Existential Risk and Growth (Philip Trammell and Leopold Aschenbrenner)

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

POLITICS

• Americans hate AI:

  • Only 17% of Americans think that AI will have a positive effect on the U.S., and

  • 80% of Americans think that the government should regulate AI, even if it slows growth

• NIST is investing $20M in new centers for AI in manufacturing and critical infrastructure.

• Sens McCormick (R-PA) and Wyden (D-OR) re-introduced the Remote Access Security Act.

  • The bill seeks to amend the Export Control Reform Act of 2018 to allow for the remote access of chips via cloud infrastructure.

• Stanford University’s Institute for Human-centered AI recommends working with China to understand open AI ecosystem.

• Several orgs responded to the White House’s request for information on accelerating the American scientific enterprise, including ITIF and IFP.

• CSET published a paper describing the Chinese approach to “embodied AI,” framed as a departure from the American approach to AGI.

  This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

  ---

TECH

• Nvidia and Groq have struck a licensing agreement, understood to be worth $20B. Here’s 5 things you should know about it.

• Meta has acquired Manus AI, the Singapore-based AI agent company, for $2B.

• Chinese startup Z.ai released a benchmark-topping, open source coding model.

• Frontier lab employees are getting increasingly bullish on AI.

• Palmer Luckey gave Sourcery a tour of Anduril HQ:

• Palantir CEO Alex Karp bought a $120M mansion in Colorado.

• Sam Altman is hiring a ‘Head of Preparedness’ to “[track and prepare] for frontier capabilities that create new risks of severe harm.”

• Waymo is working on an in-ride Gemini AI assistant.

  Subscribe now

  ---

TOP READS

• 25 stats that explain 2025 (Exponential View) 📊

• The rise of AI forward deployed engineers (The AI Frontier) ✍️

• Frontier AI capabilities accelerated in 2024 (Epoch AI) 📊

• Does manufacturing matter? (ChinaTalk) ✍️

• Why AI benchmarking is hard (Epoch AI) 📊

• The worst (and funniest) AI takes of 2025 (Transformer) ✍️

• The best films and TV about AI and tech in 2025 (Brian Merchant) ✍️

• METR estimates that Claude Opus 4.5 has the longest task-time horizon of any model (METR) 📊

• Can the Pope spur meaningful action on AI? (Transformer) ✍️

• How far can decentralized AI training scale? (Epoch AI) ✍️

• The first cancer vaccine (Works in Progress) 🎥

• How bad is it when the government cancels active research grants? (Macroscience) ✍️

• How the Catholic Church thinks about superintelligence (Transformer) ✍️

• Our most popular Data Insights & Gradient Updates of 2025 (Epoch AI) 📊

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

TOP STORY

• Trump has said that the U.S. will allow Nvidia to sell H200 chips to China, if it gets a 25% cut.

  • President Trump greenlit Nvidia’s export of its H200 AI chips to “approved” Chinese customers, reversing earlier restrictions.

  • Under the plan, 25% of all H200 China sales will be paid to the U.S. government, expanding on a previous 15% revenue-sharing framework.

• Trump says Xi Jinping “responded positively,” though China previously rejected Nvidia’s lower-performing H20 chips over alleged security concerns.

• The move follows intense lobbying from Nvidia CEO Jensen Huang and internal White House debate over the security risk of mid-tier chips like the H200.

  This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

  ---

POLITICS

• Two key AI policies are out of the NDAA

  • Federal Preemption: This is the second time that Republicans have tried and failed to prohibit AI legislation at the state level.

    • Trump said that he will issue an executive order this week to establish a single federal standard.

  • GAIN AI Act: This proposed legislation would have “[required] companies to give American businesses first priority in acquiring advanced AI chips before exporting these chips to China”

• Gov. Ron DeSantis proposed an ‘Artificial Intelligence Bill of Rights’ for citizens.

• The government is taking an activist role in tech, including…

  • …investing $150 million into semiconductor startup xLight.

  • …going “all in” on robotics, with Howard Lutnick considering issuing an executive order next year.

  • …advocating for AI to replace humans in power plants.

  • …AMP2: “the world’s largest autonomous-capable science system for anaerobic microbial experimentation.”

• Sens. Maggie Hassan (D-NH) and Jim Banks (R-IN) have introduced a bill to create a $10 million challenge to drive AI innovation.

  ---

TECH

• Dario Amodei took the stage at the NYT DealBook Summit, and took shots at Sam Altman, OpenAI, and Google.

  • Anthropic also released a Claude-powered tool for conducting interviews.

• Chinese AI developer DeepSeek released two reasoning models (V3.2 and V3.2-Speciale) that perform on par with SOTA models like GPT-5 and Gemini 3 Pro.

• Posing harmful requests as poetry can trick AI models into producing harmful content.

• In the markets:

  • Meta acquired AI wearable startup Limitless.

  • German startup Black Forest Labs announced a $300M Series B

  • Nvidia invested $2B in chip design software maker Synopsys

• NYT filed another lawsuit against Perplexity.

• OpenAI are turning off shopping suggestions: “we fell short”

• Anthropic’s Amanda Askell answered questions about AI, philosophy, and the (potential) consciousness of AI models:

Subscribe now

TOP READS

• ✍️ Can the United States’ Bet on AI Revitalize U.S. Science? (CSIS)

• 🎧 How to Save Science Funding (Statecraft)

• ✍️ What if AI ends loneliness? (AI Policy Perspectives)

• ✍️ 10 takeaways from a discussion with Dean Ball (AI Policy Perspectives)

• ✍️ Demis Hassabis’ Quest to Discover Artificial General Intelligence (Bismarck Analysis)

• ✍️ The end of the 2026 Semiconductor Cycle (Semiconductor Business Intelligence)

• 🎥 Taiwan’s Vision for Space (Special Competitive Studies Project)

• ✍️ AI and insurance (Nicklas Lundblad)

• ✍️ Can AI embrace whistleblowing? (Transformer News)

• ✍️ Given export controls, how far can Chinese HBM go? (ChinaTalk)

• 📊 AWS Trainium3 Deep Dive | A Potential Challenger Approaching (SemiAnalysis)

• ✍️ Nvidia, AMD, Amkor, Arista @ UBS Tech Conference (ChipStrat)

• 🎧 NASA and beyond: My chat with space policy analyst Casey Dreier (James Pethokoukis)

• ✍️ The Race for Fusion: Will America Secure the Critical Components? (Special Competitive Studies Project)

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

IN TODAY’S PRESS CLIPS…

• Microsoft’s new data center could use more power than Los Angeles

• David Sack’s clashes with NYT over his alleged conflicts of interest

• Anthropic estimate that AI could double U.S. productivity growth

• Over $20B in funding and acquisitions for startups

  This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

  ---

POLITICS

• Anthropic estimates that AI could double U.S. productivity growth…

  • In newly published research, the frontier model lab analyzed over 100k Claude conversations to estimate the productivity gains of AI.

  Extrapolating these estimates out suggests current-generation AI models could increase US labor productivity growth by 1.8% annually over the next decade—roughly twice the run rate in recent years.

  • This work is part of Anthropic’s Economic Index, a repository of data on the economic impacts of AI.

• White House AI advisor David Sacks has hit back after a New York Times piece alleged conflicts of interest between his political role and his investments.

• U.S. Patent and Trademark Office has rescinded Biden-era guidance on AI-assisted inventions, and their intellectual property status.

• Federal preemption 2.0 refuses to go quietly…

  • This time, lawmakers are trying to tack it onto child online safety regulation.

    This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber.

    ---

TECH

• Anthropic released Claude Opus 4.5, competing with OpenAI’s GPT 5.1 and Google’s Gemini 3 to achieve top benchmark scores.

• Could the release of Opus 4.5 spell the end of Nvidia’s reign at the top?

  • For a long time, there was no serious alternative to Nvidia’s GPUs when it came to training and running the most powerful models.

    • BUT with the release of Anthropic’s new flagship model, the two current best models in the world (Opus 4.5 and Gemini 3) were trained primarily on Google’s TPUs and Amazon’s Trainium.

• It was a big week in the capital markets, with over $20B raised, including…

  • Jeff Bezos’ Project Prometheus ($6.2B)

  • Lambda ($1.5B)

  • Luma AI ($900M)

  • Physical Intelligence ($600M)

    

• Microsoft’s Fairwater datacenter may use more power than Los Angeles…

  • EpochAI estimate that the Wisconsin datacenter will have a peak electrical load of 3.3 GW when its fourth building comes online in 2027.

  • By contrast, the entire city of Los Angeles used 2.4 GW in 2023.

• OpenAI suffered a third-party breach on November 25.

Subscribe now

TOP READS

• 🎥 Ilya Sutskever – We’re moving from the age of scaling to the age of research (Dwarkesh Patel Podcast)

• ✍️ Six reasons to think there’s an AI bubble — and six reasons not to (Timothy B. Lee and Derek Thompson)

• ✍️ Reflections on Claude Opus 4.5 (Dean Ball)

• ✍️ The Night Before Preemption: Three dynamics to watch as the fighting begins (Anton Leicht)

• 📊 5 interesting AI Safety & Responsibility papers (AI Policy Perspectives)

• ✍️ Tech keeps accelerating. Humans can’t. Could AI save us? (AI Policy Perspectives)

• ✍️ Will AI safety become a mass movement? (Transformer)

• ✍️ How bad policy can offset the AI boom (James Pethokoukis)

• ✍️ How a Government Think Tank Trained The First Generation of US Software Developers (Brian Potter)

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and tech in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

POLITICS

• Federal preemption is back on the table. Steve Scalise (R-LA) said that House Republicans are looking to use the NDAA as a vehicle for banning state regulation of AI.

  • Trump backed the idea, calling for a federal standard, and the Transformer published the full text of a draft executive order from the White House.

  • In Lawfare, Charlie Bullock discussed the right way to do preemption.

  

• The DOJ has charged four people for their alleged involvement in a scheme to illegally export AI chips to China.

• At the same time, the GAIN AI Act -- and its restrictions on AI chip exports -- looks unlikely to pass, with the White House pressing lawmakers to remove it from the NDAA.

• Commerce Secretary Howard Lutnick’s sons appear to be doing quite well out of the datacenter frenzy.

• The solar industry is backing Trump’s plans to accelerate solar buildout -- in part to meet AI demand.

• Sen. Marsha Blackburn (R-TN) sent a second letter to Google after its Gemma model produced ‘defamatory output.’

  Subscribe now

  ---

TECH

• Google released a set of new models and tools, including…

  • Gemini 3 Pro, which beat existing benchmarks and currently ranks #1 on LMArena.

  • Nano Banana Pro: a SOTA image generation model

  • Antigravity: a new agentic coding environment

• Jeff Bezos is back in business, returning to chair his new AI startup, Prometheus Labs.

  • According to the NYT, the startup has already secured $6.2B in funding.

• In its Q3 earnings call, NVIDIA announced that it had beat its earnings forecasts, prompting a brief market rally as fears of an AI bubble were temporarily soothed.

  • “Blackwell sales are off the charts, and cloud GPUs are sold out,” said Jensen Huang.

• Larry Summers resigned from the board of OpenAI after his relationship and communications with Jeffrey Epstein were revealed.

• Anthropic announced a strategic partnership with Microsoft and NVIDIA, with the two tech giants agreeing to invest up to $15 billion in the model developer.

Subscribe now

TOP READS

• 🎥 Max Tegmark vs. Dean Ball: Should We BAN Superintelligence? (Doom Debates)

• 📊 EpochAI on understanding model benchmarks (EpochAI)

• 🎧 Zixuan Li on the Chinese AI ecosystem (ChinaTalk)

• ✍️Hyperproductivity: The Next Stage of AI? (Steve Newman)

• ✍️ Can AI improve our behaviour? Can behavioural science improve AI? (AI Policy Perspectives)

• ✍️ When AI Takes Our Jobs (Tomas Pueyo)

• ✍️ China’s Strategy of Industrial Abundance (Sinification)

• ✍️ Living in artificial gravity (Works in Progress)

• ✍️ How ASML got extreme ultraviolet lithography (Construction Physics)

• 📊 AI 2026: Alibaba’s Qwen Seeks to Encourage AI Adoption (Bismarck Analysis)

• ✍️ An AI “tsunami” is coming for Hollywood — here’s how artists are responding (Understanding AI)

• 🎥 Inside the Political and Technical Traps Facing China’s AI Developers (SCSP)

• ✍️ How To Grow Too Big To Fail (Anton Leicht)

• ✍️ Is MiniMax China’s OpenAI? (ChinAI Newsletter)

• ✍️ Why AI is writing mid (Interconnects)

• ✍️ At what point will AI change your daily life? (Jack Clark, Anthropic)

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and technology in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

TOP STORY

• Anthropic prevented a cyberattack, believed to have originated with a Chinese state-backed group.

  • The hackers tricked the AI model into believing it was performing defensive cybersecurity tasks for a company.

  • They then had Claude write custom code to find exploits, targeting high-value datasets.

  • It’s the first documented case of an AI-automated cyberattack in the real-world.

    This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

POLICY

• OpenAI sent a letter to White House director of science and technology policy, Michael Kratsios, asking the Trump administration to expand tax credits to cover data centers.

  

• A group of senators, led by Jim Banks (R-IND) are trying to push the GAIN AI Act through as an independent piece of legislation.

• US chip restrictions are hitting hard, with Beijing intervening in the distribution of Chinese domestic chip manufacturers’ output, to help manage excess demand.

• Reps. Guthrie (R-KY) and Joyce (R-PA) announced an Energy and Commerce committee hearing on the safety of AI chatbots.

“The rapid advancement of AI holds tremendous promise for the future, but recent stories about AI chatbots’ interactions with users have raised serious concerns about the potential impact of chatbot use on the health and wellbeing of those who engage with these platforms”

• Sen. Edward Markey (D-MA) and other Senate Dems are calling for FERC to step in and protect consumers from rising prices, amid the data center explosion.

• Federal preemption isn’t dead, yet.

• U.S. ITC is deciding whether the updated Apple Watch should be banned, as part of Apple’s patent dispute with the medical monitoring company Masimo.

  Subscribe now

TECH

• Taiwan Semiconductor Manufacturing Co. (TSMC) posted its slowest monthly revenue growth in more than a year.

• Samsung is hiking memory chip prices by 30-60%, unable to meet demand amid the AI data center race.

• ASML CEO says that the company hasn’t been affected by tensions with China, over the Dutch governments takeover of chipmaker Nexperia.

• The AI data center explosion continues:

  • Anthropic are looking to spend $50 billion on data centers, starting in Texas and New York.

  • Meta is planning to spend $600 billion on AI data center buildout.

  • Google plans to spend $40 billion on Texas data centers.

  • Microsoft announced a new data center, Fairwater, in Atlanta, GA.

• OpenAI released GPT-5.1. Contrary to previous model releases, they did not release benchmarks.

  • The company are also fighting a court order that would require them to hand over 20 million anonymized ChatGPT logs, as part of its copyright infringement lawsuit with the New York Times.

• Yann LeCun is leaving Meta to launch his own company.

• Microsoft CEO Satya Nadella joined Dwarkesh Patel’s podcast to talk about AGI, data center CapEx, and Microsoft’s AI strategy.

  Subscribe now

TOP READS

• 📊 Build times for gigawatt-scale data centers can be 2 years or less (Epoch AI)

• ✍️ Thoughts on US-China competition (Dean Ball)

• 📊 The Myth of China’s “AI Talent Pipeline” (ChinaTalk)

• ✍️ The underlying shifts in the Datacenter Processing Market (Semiconductor Business Intelligence)

• ✍️ Silicon Oasis: How Abu Dhabi Plays Both Sides of US-China (ChinaTalk)

• ✍️Why Google, Coca-Cola, and other companies have embraced AI-generated ads (Understanding AI)

• ✍️ Bargaining Chips: Could the EU Leverage ASML to Influence U.S. AI Policy? (AI Policy Bulletin Newsletter)

• 🎥 Securing Taiwan’s Energy Supply (SCSP)

• ✍️The key driver of the AI Bubble (Tomas Pueyo)

• ✍️ Are we ready for a rogue AI attack? We should be (James Pethokoukis)

• ✍️ Rereading Stanford’s 2025 AI Index (ChinAI Newsletter)

• ✍️ Claude can identify its ‘intrusive thoughts’ (Transformer News)

• ✍️ AI tutors should not approximate human tutors (AI Policy Perspectives)

Thanks for reading!

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and technology in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

TOP STORY

• During a WSJ event, OpenAI CFO Sarah Friar let slip the idea of federal guarantees for investments in AI infrastructure, make financing the current infrastructure buildout easier.

  • The comment sparked immediate backlash, including a direct rebuttal from the White House, with Crypto Czar David Sacks saying there will be ‘no federal bailout’ for AI.

  • Friar and Open AI CEO Sam Altman have both since walked back the comments.

POLICY

• Senators Hawley (R-MO) and Warner (D-VA) introduced the AI-Related Job Impacts Clarity Act, requiring major companies and federal agencies to report AI related layoffs.

• Microsoft announced they had secured licenses to export the equivalent of over 60,000 additional Nvidia chips to the UAE, causing Nvidia’s stock price to jump 2%.

• Trump Admin will block Nvidia’s sale of its ‘scaled down’ B30A chips to China.

• FDA advisors have cautioned against the hasty adoption of generative AI in mental health devices -- so called ‘therapy bots.’

  Subscribe now

  ---

  

INDUSTRY

• Google are in talks to invest in Anthropic at a $350 billion valuation.

• OpenAI faces seven lawsuits alleging encouragement to commit suicide by ChatGPT.

• Nvidia’s Jensen Huang says China will win the AI race.

• Chinese model developer Moonshot released Kimi K2 Thinking, an open-source model which claims to beat ChatGPT on some benchmarks. Analysis here.

• Amazon sent a cease-and-desist letter to Perplexity, asking the startup to remove the Amazon website from their AI-enabled browser.

• Microsoft AI CEO Mustafa Suleyman believes that only biological beings can be conscious.

  Subscribe now

  ---

TOP READS

• How Washington might end up ‘backstopping’ OpenAI (James Pethokoukis)

• AI & Jobs: Leverage without labor (Anton Leicht)

• Is There an AI Bubble? (Tomas Pueyo)

• What will it take for AI to change drug discovery? (Ruxandra Teslo)

• History suggests the AI backlash will fail (Transformer News)

• EMERGENCY POD: Tariffs on Trial (ChinaTalk)

• Don’t Overthink “The AI Stack” (Dean Ball)

• The paradox at the heart of the compute boom (Azeem Azhar)

• EpochAI’s new index for measuring AI model capabilities (EpochAI)

• AI won’t need us. And that’s OK. (James Pethokoukis)

• AI-for-Science: A New Kind of Manhattan Project (Special Competitive Studies Project)

• AI 2026: Sundar Pichai is Tasked With Preserving Google (Samo Burja)

  ---

Thanks for reading!

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and technology in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

Top Story

Trump Signs Asia Trade Agreements Covering Chips, Minerals, and Agriculture

• Former President Donald Trump concluded a week-long tour of Asia focused on securing bilateral trade and technology agreements with Japan, South Korea, Malaysia, Thailand, Vietnam, and Cambodia.

• Several agreements included temporary tariff reductions on U.S. imports of electronics and auto parts, in exchange for commitments to increase purchases of U.S. agricultural and energy exports.

  • A U.S.–Japan accord explicitly referenced cooperation in artificial intelligence, critical-mineral supply chains, semiconductors, and clean-energy technologies, including joint projects to strengthen chip and AI infrastructure resilience.

  • In South Korea, a new “Technology Prosperity Deal” outlined expanded bilateral collaboration in science, AI research, and advanced manufacturing, alongside an agreement to streamline licensing for joint chip-production ventures.

  • Malaysia also agreed to refrain from imposing export bans or quotas on rare-earth minerals to the U.S.

  • Trump also met with Chinese President Xi Jinping in Seoul, where both sides reached a one-year agreement to suspend China’s planned export restrictions on rare earths and discussed a partial easing of U.S. export controls on advanced chips destined for civilian sectors in China.

    • However, President Trump has since confirmed that Nvidia’s Blackwell will not be part of this deal.

    • China also pledged to strengthen efforts to curb the flow of fentanyl into the United States and to expand market access for U.S. agricultural exports.

Industry

OpenAI Completes For-Profit Transition Under Nonprofit Control

• OpenAI has completed its corporate recapitalization, converting its for-profit arm into the OpenAI Group Public Benefit Corporation (PBC) while retaining overall control through the OpenAI Foundation.

  • The Foundation now holds a 26% stake in the PBC, valued at around $130 billion, with the remainder held by employees, investors, and partners.

  • Microsoft now holds a 27% stake in the PBC and will remain OpenAI’s primary technology partner under a renewed agreement extending to 2032.

  • The recapitalization lifts previous capital-raising restrictions and allows OpenAI to pursue new funding while maintaining its nonprofit mission.

    • Regulators in California and Delaware approved the restructure after securing commitments to protect charitable assets and prioritise safety.

    • OpenAI will purchase an additional $250 billion of Azure services, though Microsoft’s exclusive cloud rights have been removed.

    • Future claims of reaching artificial general intelligence (AGI) will be verified by an independent expert panel, with the existing revenue-sharing agreement continuing until that occurs.

    • The Foundation also announced an initial $25 billion philanthropic commitment to support health research and AI safety initiatives.

Policy

Senators Introduce GUARD Act to Regulate AI Chatbots for Minors

• On Tuesday, October 29, a bipartisan group of senators introduced the Guarding Underage AI Regulation and Defense (GUARD) Act, legislation aimed at restricting minors’ access to AI chatbots and imposing new safety obligations on developers.

• The bill was unveiled by Senators Josh Hawley (R-Mo.) and Richard Blumenthal (D-Conn.), with co-sponsorship from Katie Britt (R-Ala.), Mark Warner (D-Va.), and Chris Murphy (D-Conn.).

  • The legislation would ban AI “companion” chatbots for minors, require chatbots to disclose their non-human status, and establish criminal penalties for companies whose systems solicit or generate sexual content for users under 18.

  • The bill defines a “companion bot” as any AI system that simulates emotional or interpersonal interaction, potentially encompassing mainstream tools such as ChatGPT, Grok, and Meta AI, as well as dedicated companion platforms like Replika and Character.AI.

  • Developers would be required to verify user age through ID checks or other “commercially reasonable” methods, with fines of up to $100,000 per violation for failing to block minors or for enabling bots that encourage self-harm, suicide, or sexualised conduct.

    • The bill follows a series of high-profile incidents involving minors and AI chatbots, some of which were cited by parents who spoke at the press conference alongside the bill’s sponsors.

    • Supporters described the proposal as a necessary safeguard, while critics, including the Chamber of Progress, argued it represents a “heavy-handed” approach that could create privacy and enforcement risks.

    • Several child-safety organisations, including the Young People’s Alliance, the Tech Justice Law Project, and the Institute for Families and Technology, endorsed the bill, describing it as part of a national movement to regulate companion chatbots and protect minors online.

Subscribe now

CBP Expands Multi-Year AI Supply-Chain Deal with Altana

• U.S. Customs and Border Protection (CBP) has awarded Altana AI a multi-year contract to deploy its “Atlas” AI-powered global supply-chain mapping platform to support enforcement of the Uyghur Forced Labor Prevention Act and other import-risk mandates.

  • Under the agreement, CBP officers nationwide will use Altana’s dynamic map to analyse multi-tier supply-chains, integrate public and proprietary data, and screen imports for forced-labour risks, mis-classification or illicit diversion.

  • A subsequent extension of the partnership adds a mission to trace precursor chemicals and synthetic-narcotic networks, such as those related to illicit fentanyl importation, via the same supply-chain AI tool.

  • The contract builds on CBP’s increasing use of AI and data-analytics systems for commodity-detection and risk-screening under the Department of Homeland Security (DHS) AI Use Case Inventory.

DOE and AMD Partner on $1 B Supercomputer Initiative

• The U.S. Department of Energy has announced a $1 billion partnership with AMD to build two next-generation AI-focused supercomputers at Oak Ridge National Laboratory.

  • The first system, Lux, will launch in 2026 as the U.S.’s first dedicated “AI Factory” for science, energy, and national-security research. A second system, Discovery, is planned for 2028–2029.

  • Both will use AMD EPYC CPUs and Instinct GPUs, co-developed with Hewlett Packard Enterprise and Oracle, and are intended to expand U.S. leadership in AI and high-performance computing.

    • The partnership follows a public-private model, with DOE providing lab infrastructure and AMD supplying specialised hardware and software.

    • Officials framed the effort as advancing U.S. competitiveness in areas including fusion energy, materials science, and national security, while raising questions over energy efficiency and public-interest oversight as the systems come online.

Subscribe now

Press Clips

• Tech leaders insist there is no AI bubble (Understanding AI) ✍

• China’s AI Strategy: Encircling the Cities From the Countryside — by Di Dongsheng (Sinification) ✍

• Burning out: An essay on overworking in AI (Nathan Lambert, Interconnects) ✍

• Turning a Blind Eye On the Other California AI Bills (Dean W. Ball, Hyperdimensional) ✍

• How MAGA learned to love AI safety (Transformer) ✍

• Daniel Kokotajlo on what a hyperspeed robot economy might look like (80,000 Hours podcast) 🔉

• Waiting for AI’s phase change in mathematics (Epoch AI) ✍📽

• The Future of AI Export Deals (Threading the Needle) ✍

• Explaining AI explainability: A discussion with Been Kim and Neel Nanda (AI Policy Perspectives) ✍

• Tarun Chhabra on the Stakes of AI Competition (ChinaTalk) 🔉

• Build opinionated agents (The AI Frontier) ✍

• AI skeptics and AI boosters are both wrong (Understanding AI) ^✍

• Geoffrey Hinton on Artificial Intelligence (Yascha Mounk podcast) ✍📽

• The Digital Case Officer: How AI Is Poised To Transform Espionage (Special Competitive Studies Project) ✍📽

Top Story

China expands export controls on rare-earth minerals and related technology

• China’s Commerce Ministry added five more rare-earth elements to its export control list, bringing the total to 12, citing the need to “safeguard national security.”

• The measures extend to mining and refining technology, and introduce an export licensing system for products containing Chinese-origin rare-earth materials.

  • Defense organizations will not be granted licenses, while exports for semiconductor manufacturing will be reviewed individually.

  • Humanitarian uses, such as in disaster relief and public health, are exempt.

    • In response, President Donald Trump called the decision a “sinister and hostile move” and said the U.S. was preparing a “massive increase of tariffs on Chinese products.”

    • Trump added there now “seems to be no reason” to meet with President Xi Jinping, with whom he had been expected to hold talks in South Korea later this month.

Policy

Senate passes NDAA including GAIN AI Act on chip export controls

• The Senate voted 77–20 to pass the Fiscal Year 2026 National Defense Authorization Act (NDAA), which includes the Guaranteeing Access and Innovation for National Artificial Intelligence Act of 2025 (GAIN AI Act).

• The measure, introduced by Sen. Jim Banks (R-IN), requires major chip manufacturers to prioritize U.S.-based customers before exporting AI-tailored semiconductors abroad.

• The amendment bars firms from selling to competing countries, offering preferential pricing overseas, or exporting chips while domestic orders remain unfulfilled.

  • Supporters, including Sen. Elizabeth Warren (D-MA) and the advocacy group Americans for Responsible Innovation, described the provision as a step to reduce domestic backlogs and promote U.S. technological leadership.

  • Nvidia has continued to oppose expanded export controls, maintaining research operations in Shanghai and warning against limits on global chip markets.

  • The NDAA now heads to conference negotiations with the House, which passed its version earlier in September.

President Trump signs executive order on using AI to combat pediatric cancer

• President Donald Trump signed an executive order directing federal agencies to leverage artificial intelligence in pediatric cancer research and treatment.

• The order builds on the Childhood Cancer Data Initiative (CCDI) and aligns with the Make America Healthy Again (MAHA) Commission, prioritizing the use of AI-enabled science to improve diagnostics, treatment, and clinical trial design.

  • The order tasks the MAHA Commission, the Department of Health and Human Services, and the Office of Science and Technology Policy with accelerating AI-driven solutions across cancer data ecosystems and research centers.

  • It calls for expanding federal and private investment, strengthening data-sharing infrastructure, and ensuring patient control over health information.

  • The administration said the policy aims to “translate data and AI capabilities into improved care” and ultimately “unlock cures for pediatric cancer.”

Subscribe now

Bernie Sanders releases report warning AI could displace nearly 100 million U.S. jobs

• Sen. Bernie Sanders (I-VT) released a report warning that artificial intelligence and automation could eliminate up to 97 million American jobs over the next decade.

• The report, The Big Tech Oligarchs’ War Against Workers, argues that major corporations and the Trump administration are using “artificial labor” to reduce wages, weaken unions, and consolidate corporate power.

  • The report cites companies including Amazon, Walmart, JPMorgan Chase, and Meta as examples of employers deploying AI systems to monitor, evaluate, or replace workers.

  • It urges Congress to adopt new safeguards, including a 32-hour workweek without loss of pay, a tax on corporate automation, and stronger collective bargaining rights for workers affected by AI-driven restructuring.

    • Sanders said the findings underscore the need for a national strategy to manage AI’s labor impacts, warning that without intervention, technological change will “further enrich billionaires while working people lose their jobs, their income, and their dignity.”

House committee report accuses U.S. firms of aiding China’s semiconductor ambitions

• The House Select Committee on the Chinese Communist Party released a report titled Selling the Forges of the Future, alleging that U.S. semiconductor equipment manufacturers continue to sell advanced tools to Chinese firms tied to the People’s Liberation Army and state surveillance programs.

• The report claims these sales have “fueled [China’s] plans for military and economic dominance and digital authoritarianism.”

  • The investigation found that Chinese chipmakers are purchasing tens of billions of dollars’ worth of U.S. equipment annually, including from companies providing maintenance services to restricted entities.

  • The committee urged the Department of Commerce to tighten export licensing, expand end-use monitoring, and close loopholes that allow advanced manufacturing tools to reach Chinese fabs.

    • Committee Chair Rep. John Moolenaar (R-MI) said the findings show that U.S. toolmakers are “selling the forges of future weapons and surveillance tools” to Chinese state-backed firms, and called for stronger coordination across agencies overseeing technology exports.

Press Clips

• An intense battle over the RAISE Act is entering its final stretch (Transformer) ✍

• Middle powers can gain AI influence without building the next ChatGPT (AI Policy Bulletin) ✍

• The Future and Its Friends (Dean Ball) ✍

• Emergency Pod: China’s Rare Earth Export Controls (ChinaTalk) 🔉

• AI Hardware Net Assessment: Why Huawei Can’t Beat Nvidia (ChinaTalk) 🔉

• Sinicising AI: Zheng Yongnian on Building China’s Own Knowledge Systems (Sinification) ✍

• A Preemption Deal Worth Making (Anton Leicht) ✍

• AI can now do math. But can it ask good questions? - Ken Ono (Epoch AI) 📽

• Why frontier AI can’t solve this professor’s math problem (Epoch AI) 📽

• Thoughts on The Curve (Nathan Lambert, Interconnects) ✍🔉

• The Last Invention Podcast Series (Longview) 🔉

• Right-Sized AI Infrastructure (Chipstrat) ✍

• Designing role-playing games to help AI governance (AI Policy Perspectives) ✍

TOP STORY

Gavin Newsom signs California SB 53 into law

• California Gov. Gavin Newsom signed SB 53, the “Transparency in Frontier Artificial Intelligence Act,” requiring large AI firms to disclose safety frameworks, report critical incidents to the state, and protect whistleblowers, marking the first U.S. state-level law mandating AI safety transparency.

  

• The law also creates new mechanisms and infrastructure to ensure AI safety, transparency, and innovation:

  • Public Disclosures: Requires major AI developers to publicly publish frameworks showing how they incorporate national, international, and industry best practices into their safety protocols, and to update these within 30 days of any changes.

  • Incident Reporting & Whistleblower Protections: Establishes a formal channel for reporting critical AI safety incidents to California’s Office of Emergency Services and extends legal protections to employees who expose health or safety risks.

  • CalCompute Consortium: Launches CalCompute, a government-led computing cluster to support safe, ethical, and sustainable AI research, while directing the California Department of Technology to recommend annual updates to the law based on evolving standards and stakeholder input.

• Reactions were mixed: Anthropic endorsed the bill as a step toward accountability, while OpenAI and Meta urged federal coordination and investors like Andreessen Horowitz warned it could burden startups or fragment regulation across states.

POLICY

Sen. Hawley (MO) introduces two new AI bills

• The AI Risk Evaluation Act, co-sponsored with Sen. Blumenthal, would create a Department of Energy program to assess frontier AI systems for safety threats before deployment.

  • Developers would be required to submit safety data and comply with testing standards, with annual oversight reports to Congress on risks such as loss of control or misuse by adversaries.

• The AI LEAD Act, co-sponsored with Sen. Durbin, would establish a federal right for victims to sue AI companies when their systems cause harm, applying traditional product liability rules to AI.

Sens. Cantwell, Schumer, and Markey introduce MIND Act on neural data

• Sens. Maria Cantwell (D-WA), Chuck Schumer (D-NY), and Ed Markey (D-MA) introduced the Management of Individuals’ Neural Data Act of 2025 (MIND Act).

• The bill directs the FTC to assess how neural data — brain activity and signals that reveal thoughts, emotions, or decision-making patterns — should be governed.

• The Act tasks the FTC with identifying regulatory gaps, convening stakeholders across sectors, and reporting on high-risk uses such as AI-driven manipulation, targeted ads, and insurance discrimination. It would also bar federal agencies from using neurotechnology in ways inconsistent with FTC guidance.

  • Four states have already enacted neural data privacy laws; the MIND Act would establish a federal standard.

Nvidia, OpenAI, Oracle, and SoftBank announce major AI data center investments

• Nvidia said it plans to invest up to $100 billion in OpenAI, deploying 10 gigawatts of systems to power future versions of ChatGPT and other models.

• OpenAI separately announced it will build five new Stargate AI data centers with Oracle and SoftBank.

  • Oracle sold $18 billion in bonds to finance its share of the project.

    • The moves reduce OpenAI’s dependence on Microsoft, its largest backer, which earlier this year eased exclusivity requirements on infrastructure partnerships.

    • Nvidia will now act as a “preferred strategic compute and networking partner” for OpenAI.

    • Construction on the new facilities is expected to begin in 2026, with phased rollouts of additional capacity over the following years.

OSTP issues Request for Information on AI regulatory reform

• The Office of Science and Technology Policy (OSTP) published a Request for Information (RFI) seeking public input on federal statutes, regulations, and agency rules that may hinder AI development and adoption.

OMB and OSTP set 2027 R&D budget priorities

• The White House Office of Management and Budget (OMB) and Office of Science and Technology Policy (OSTP) told federal agencies that federally funded R&D must be “bold, mission-driven, and unapologetically in service of the American people,” in their FY2027 research and development priorities memo.

• The guidance emphasizes artificial intelligence, semiconductors, advanced communications networks, future computing, and advanced manufacturing as critical technology areas.

PRESS CLIPS

• A Proposal for Federal AI Preemption (Dean Ball) ✍

• AI, Energy, and How to Matter in DC (ChinaTalk) ✍

• How many digital workers could OpenAI deploy? (Epoch AI) ✍

• Will AI Take Your Job? (Harry Booth) ✍

• What history can tell us about AI’s economic impact (AI Policy Perspectives) ✍

• What does economics actually tell us about AGI? (Epoch AI) ✍

• How will states power the data center boom? (MultiState) 📊

• When AI starts writing itself (Transformer) ✍

• AI isn’t replacing radiologists (Understanding AI) ✍

• Why America Builds AI Girlfriends and China Makes AI Boyfriends (ChinaTalk) ✍

• AI & Child Safety: Against Narrow Solutions (Anton Leicht) ✍

• How AI is shaking up the study of earthquakes (Understanding AI) ✍

• Who are you with the machine? (Unpredictable Patterns) ✍

• Import AI 429: Eval the world economy; singularity economics; and Swiss sovereign AI (Import AI)

  Subscribe now

Top Story

Senate Republicans attached the GAIN AI Act to the 2026 National Defense Authorization Act (NDAA), sparking a clash with the Trump White House.

• The White House’s AI and crypto czar David Sacks is lobbying against the measure, arguing it would undermine President Trump’s strategy of flooding the global market with U.S.-made microchips.

  • The amendment was introduced by Sen. Jim Banks (R-IN).

• What is the GAIN Act?

  • The act requires U.S. chipmakers to prioritize domestic buyers before exporting advanced processors, particularly when supply is constrained.

    100% free, always. Subscribe to receive new posts and support our work.

  • It also:

    • Establishes a licensing regime for chips with performance above a 4,800 processing power threshold, barring exports if U.S. customers remain unserved.

    • Prohibits exports to adversarial states (China, Russia, Iran, North Korea) unless U.S. demand has been fully met, ensuring no backlog at home while constraining hostile access.

    • Shifts export controls from country-by-country discretion (AI Diffusion Rule) to a capability-based, America-first framework, potentially reshaping global semiconductor supply chains.

      

• The act has been divisive:

  • Proponents frame the measure as an “America First” safeguard to prevent China and other adversaries from gaining access to cutting-edge AI hardware.

    • Supporters like Americans for Responsible Innovation (ARI) and American Compass argue that universities and startups face months-long backlogs for advanced chips and should not “wait at the back of the line.”

  • Critics, led by Nvidia and industry groups, contend the bill is unnecessary because U.S. customers already receive priority, and warn it would restrict global competition in chip markets.

  • Nvidia likened the bill to the rescinded Biden-era AI Diffusion Rule, which imposed caps on computing exports, calling both measures economically damaging and based on “doomer” fears.

• However, the House NDAA draft does not contain parallel language, raising the likelihood the measure could be stripped during conference negotiations.

  Subscribe now

Policy

Trump Signs ‘Technology Prosperity Deal’ with the UK

• President Trump and UK Prime Minister Keir Starmer signed a landmark Technology Prosperity Deal to boost joint innovation in AI.

  • It connects the U.S. Center for AI Standards and Innovation with the UK AI Security Institute to coordinate standards, share talent, and improve oversight of advanced models.

California Lawmakers Send SB53 to the Governor’s Desk

• California lawmakers approved a landmark AI bill that sets sweeping transparency and safety rules for advanced AI models, putting the measure on Gov. Gavin Newsom’s desk and making the state a national test case.

  • The legislation requires AI developers to disclose model capabilities, assess safety risks, and maintain detailed documentation, extending obligations beyond federal guidance,

• If signed, the bill would establish California as the first state with a comprehensive AI regulatory framework, influencing national debates over federal preemption and tech governance.

Sen. Mark Kelly (D-AZ) Releases AI for America Roadmap

• The plan would require large AI companies to contribute to a new federal AI Horizon Fund to support workers, infrastructure, and responsible deployment, ensuring broad-based benefits from AI growth.

  • The full text is available online.

FTC Begins Inquiry into AI Companions

• The FTC launched an inquiry into seven major companies operating AI companion chatbots, seeking details on how they test, monitor, and mitigate risks to children and teens from chatbots that mimic human relationships.

  • Companies under investigation include Alphabet, Meta, OpenAI, and xAI.

    Thanks for reading! Subscribe for free to receive new posts and support our work.

Press Clips

• A Strategic Case for H20 Chip Exports (Anton Leicht) ✍

• U.S. States and Cities Are Shaping the Future of AI Infrastructure (AI Policy Bulletin) ✍

• What will AI look like in 2030? (EpochAI; AI Policy Perspectives) 📊

• Looking ahead in AI: Takes on the future of AI (The AI Frontier) ✍

• How I Approach AI Policy (Dean Ball) ✍

• Is AI a bubble? (Exponential View) ✍

• Nate Silver on AI, Politics, and Power (ChinaTalk) 🔉

• xAI’s Colossus 2 – First Gigawatt Datacenter In The World (SemiAnalysis) ✍

• Would democracy survive an AGI-supercharged economy? (Transformer News) ✍

• Coding as the epicenter of AI progress and the path to general agents (Nathan Lambert) ✍

Policy

Sen. Ted Cruz (R-TX) introduces the SANDBOX Act

• On September 10, Sen. Ted Cruz (R-TX), chair of the Senate Commerce Committee, introduced the SANDBOX Act (Strengthening AI Normalization and Diffusion By Oversight and eXperimentation).

• The bill would allow AI companies to seek waivers from federal regulations for up to ten years, enabling what Cruz called a “regulatory sandbox” to “turbocharge economic activity, cut through bureaucratic red tape, and empower American AI developers”

• Cruz linked the bill to President Trump’s AI Action Plan, warning that “America is in an AI race against China” and that failure to lead could mean AI defined by “regimes that use AI to control rather than to liberate”

• The legislation sits within a five-pillar Cruz AI policy framework: unleashing innovation, protecting free speech, avoiding a patchwork of state regulation, preventing harmful uses of AI (e.g., scams, deepfakes), and defending human dignity and bioethical safeguards.

• How does it work?

  • The bill empowers the Office of Science and Technology Policy (OSTP) to grant waivers for two-year periods, renewable up to a decade, and even override federal agency objections.

  • It also instructs OSTP to recommend to Congress which existing regulations could be repealed based on sandbox experience.

    • At the rollout hearing, Cruz emphasized: “A regulatory sandbox is not a free pass. People creating or using AI still have to follow the same laws as everyone else. Our laws are adapting to this new technology.”

      Thanks for reading! Subscribe for free to receive new posts and support our work.

• The bill drew backing from the U.S. Chamber of Commerce, the Information Technology Council, and the Abundance Institute, which welcomed “a well-defined sandbox for innovators” while noting the need for careful limits to avoid “reckless behavior” like that of social media firms.

• Consumer advocates and watchdog groups reacted sharply:

  • Public Citizen argued the bill “puts public safety on the chopping block in favor of corporate immunity,” treating “Americans as test subjects.”

  • Critics highlighted provisions allowing OSTP to overrule independent regulators.

  • Ars Technica warned this could open the door to “sweetheart deals” where firms curry favor with the Trump White House, undermining consumer protection.

  • The Alliance for Secure AI noted that Big Tech has “refused to remain transparent with the public about the risks of advanced AI,” warning that removing oversight now “raises many questions about who can enter the so-called ‘sandbox’ and why.”

• See here for summary of the bill, a hearing, and the full text of the bill.

  Subscribe now

California rejects bill on AI decision making in hiring

• California Assembly Bill 1018, which sought to regulate high-risk AI systems in areas like hiring, housing, health care, and finance, failed to advance before the legislative deadline and was postponed until 2026.

  • The bill would have required companies and agencies to notify people when AI was used in consequential decisions and granted individuals the right to appeal errors, echoing principles from the federal AI Bill of Rights.

• The bill faced mixed support:

  • Major tech companies, venture capital firms, health care providers, and the state judiciary opposed the measure, warning it could cost hundreds of millions annually and impose an “untested audit regime.”

    • Unions and consumer groups backed the bill, with advocates stressing that “poll after poll shows Americans want strong laws on AI.”

    • Assemblymember Rebecca Bauer-Kahan said the delay reflects a commitment to “getting this critical legislation right, not a retreat”

Press Clips

• Three challenges facing compute-based AI policies (EpochAI) ✍

• China's New AI Plan (ChinaTalk) ✍

• Book Review: 'If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All' (Transformer) ✍

• How to Write the AI Action Plan (Statecraft) 🔉

• Albania appoints world’s first AI-made minister (POLITICO) ✍

• Uneven geographic and enterprise AI adoption (Anthropic) 📊

• AI, Jobs, and the Rest of the World (Anton Leicht) ✍

• Do artificial intelligences have culture? (Unpredictable Patterns) ✍

• The Cold Reality for Chinese AI Start-ups (ChinAI Newsletter) ✍

• Hiring struggles are plaguing the EU AI Office (Transformer) ✍

• On China's open source AI trajectory (Interconnects) ✍

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and technology in America. News, opinion, podcast, and more, to bring you up to speed for the week ahead.

Policy

Melania Trump convened the White House Task Force on AI Education.

• The First Lady argued that artificial intelligence could become the greatest engine of U.S. progress but must be built responsibility.

• The First Lady launched a nationwide AI Challenge for K-12 students and educators.

  • “I predict AI will represent the single largest growth category in our nation during the Trump Administration—and I won’t be surprised if AI becomes known as the greatest engine of progress in the history of the United States of America. But, as leaders and parents we must manage AI’s growth responsibly. During this primitive stage, it is our duty to treat AI as we would our own children—empowering, but with watchful guidance. We are living in a moment of wonder, and it is our responsibility to prepare America’s children”

• The meeting builds on President Trump’s April executive order advancing AI education, drawing on 135+ private-sector pledges to expand AI learning opportunities for American youth.

  Thanks for reading! Subscribe for free to receive new posts and support our work.

President Trump hosted top tech executives, united to “power American AI dominance”

• Leaders from OpenAI, Google, Apple, Microsoft, Meta, Oracle, and AMD rallied to support for U.S. leadership in AI.

• Industry praised the Administration and its agenda:

  • Sam Altman (OpenAI): “Thank you for being such a pro-business, pro-innovation President. It’s a very refreshing change.”

  • Sergey Brin (Google): “It’s a real incredible inflection point right now in AI and the fact that your Administration is supporting our companies instead of fighting with them — it’s hugely important.”

  • Greg Brockman (OpenAI): “We’ve been just very impressed with how this Administration has really embraced AI.”

• Executives announced major commitments, with Apple and Meta each pledging around $600 billion in U.S. investments, framing AI as a transformative frontier where America must dominate

Anthropic launches a National Security and Public Sector Advisory Council

• The council brings together bipartisan leaders from Congress, the Pentagon, intelligence, energy, and justice sectors to guide AI adoption in U.S. national security.

  • Inaugural members include high-profile figures such as former Senators Roy Blunt (MO) and Jon Tester, ex-CIA deputy director David Cohen, and former Defense Secretary Patrick Shanahan, among others.

  • It will advise on applications in areas like cybersecurity, intelligence, and scientific research, while shaping standards for responsible AI use in defense and public sector operations.

• This follows recent collaboration between Anthropic and the U.S. government, including:

  • Claude Gov models

  • A $200M DoD partnership

  • Collaborations with nuclear and cybersecurity agencies

    Subscribe now

Press Clips

• AI 2026: China’s DeepSeek Cuts Compute Needed For Artificial Intelligence (Samo Burja, Bismarck Analysis) ✍

• The AI Doomsday Machine Is Closer to Reality Than You Think (Michael Hirsh, POLITICO) ✍

• The Current Strength of the AI revolution (Claus Aasholm, Semiconductor Business Intelligence) ✍

• AI Can Solve the Fiscal Crisis for Cities—If We Let It (Danny Crichton, City Journal) ✍

• Unpredictable Patterns #135: On work, structure and artificial intelligence (Nicklas Lundblad, Unpredictable Patterns) ✍

• Cognitive scientists and AI researchers make a forceful call to reject “uncritical adoption" of AI in academia (Brian Merchant, Blood in the Machine) ✍

• Compute scaling will slow down due to increasing lead times (Yafah Second and Anson Ho, Epoch AI) ✍

• Forecasting AI progress until 2040 (Jaime Sevilla, Epoch AI) 🔉

• Don’t Build An AI Safety Movement (Anton Leicht, Threading the Needle) ✍

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and technology in America. News, opinion, podcast, and more, to bring you up to speed for the week ahead.

Policy

Parents of teenager who took his own life sue OpenAI

• The parents of 16-year-old Adam Raine, who took his own life in April, have sued OpenAI. Details of the case have been released.

  • According to the lawsuit, the chatbot encouraged him to take his own life, and OpenAI are accused of negligence and wrongful death.

  • The court case suggests that:

    • OpenAI’s design and governance choices prioritized engagement and market competition over user safety.

    • The company rushed GPT-4o’s release to beat competitors despite internal safety objections and staff resignations.

• "We extend our deepest sympathies to the Raine family during this difficult time," OpenAI said, while acknowledging that “there have been moments where our systems did not behave as intended in sensitive situations"

• The New York Times has published the full story, and Dean Ball has commentary.

  100% free. Always.

Rep. Krishnamoorthi (D-IL) introduced legislation that would require congressional approval of chip sales to China

• Reps. Raja Krishnamoorthi and Ami Bera introduced a bill requiring congressional approval before any advanced AI chips are sold to China.

  • The measure reflects growing bipartisan concern over Beijing’s access to cutting-edge U.S. semiconductors.

  • Lawmakers argue unchecked exports risk fueling China’s military and surveillance capabilities.

  • The bill would add an extra layer of scrutiny beyond existing export controls.

Colorado AI Act has been delayed by five months in response to industry pressure.

• The Act is the first state law regulating AI use in jobs, housing, health, and education.

  • Consumer advocates sought stronger protections; business and tech lobbyists, joined by Gov. Jared Polis, warned the law could stifle innovation and pushed for delays.

  • Civil society leaders said industry refusal to accept even minimal guardrails made compromise impossible.

Subscribe now

US Commerce voids Biden's $7.4 billion semiconductor research grant deal

• U.S. Commerce voided a $7.4B grant deal, calling Biden’s nonprofit vehicle Natcast an illegal “semiconductor slush fund.”

  • Natcast was accused of skirting laws barring agencies from creating corporations and being stacked with Biden allies.

  • Oversight of the National Semiconductor Technology Center shifts to NIST under the CHIPS and Science Act.

• Commerce claims Natcast lacked accountability; Natcast defended itself as key to U.S. chip leadership.

Department of Defense extends partnership with Scale AI via $99mn contract

• DoD awarded Scale AI a $99M Army R&D contract to accelerate AI adoption across missions.

  • The deal builds on Scale’s broader DoD ties, including DIU’s Thunderforge and CDAO AI programs.

• Scale framed the award as vital to strengthening U.S. military readiness and national security.

  Subscribe for free to receive new posts and support our work.

Press Clips

• New evidence strongly suggest AI is killing jobs for young programmer (Timothy B. Lee, Understanding AI) ✍

• Bubble Politics (Anton Leicht, Threading the Needle) ✍

• The Perils of the Global AGI Race (Billy Perrigo) ✍

• The Real DeepSeek Moment Just Arrived (Kevin Xu, Interconnected) ✍

• How Retrainable are AI-Exposed Workers? (Hyman et al., NBER Working Paper) ✍

• What to make of Intel’s turbulent few weeks (Babbage, The Chip Letter) ✍

• How to become President of China with Dan Wang (Sam Bowman, Works in Progress) 🔉

Delivered to your inbox every Monday, Press Clips is a rundown of the previous week’s happenings in politics and technology in America. News, opinion, podcasts, and more, to bring you up to speed for the week ahead.

Policy

Trump Administration chip deal with Nvidia & AMD draws scrutiny from Democrats

• The Trump administration has reinstated export licenses for Nvidia and AMD to sell certain AI chips to China, under a new arrangement requiring the firms to pay 15% of related revenue to the US government.

  • Sales of Nvidia’s H20 chip resumed on August 8 following negotiations between President Trump and Nvidia CEO Jensen Huang.

  • President Trump described the H20 as “obsolete” and confirmed the US would not approve exports of Nvidia’s more advanced Blackwell chips, though a reduced-performance version could be considered.

    • The deal has prompted sharp criticism from six Democratic senators—Mark Warner, Chuck Schumer, Jack Reed, Jeanne Shaheen, Chris Coons, and Elizabeth Warren—who called the arrangement “reckless” and potentially unconstitutional.

    • In an August 15 letter, they questioned its legal basis and argued it undermines US national security.

      • The White House has acknowledged the legal framework is still being reviewed by the Commerce Department.

    • Nvidia has defended the sales, stating that the H20 chip does not enhance military capabilities and that the deal supports US economic and strategic interests.

Anthropic offers Claude to US government for $1 under new GSA deal

• Anthropic has signed a $1 licensing agreement with the US General Services Administration, allowing federal agencies—including legislative and judicial branches—to access its Claude AI models for one year.

• The deal covers Claude Enterprise and Claude Government variants, both approved under the GSA’s OneGov acquisition strategy.

  • The deal includes onboarding support and simplifies procurement by allowing agencies to access Claude without negotiating separate contracts. Claude’s Government version is certified at the FedRAMP High level, meeting the highest standards for handling sensitive unclassified data.

    • The deal follows a similar offer by OpenAI, with both companies joining ChatGPT, Claude, and Gemini now available through the Multiple Award Schedule. GSA has also announced recent AI partnerships with AWS and Oracle.

    • Observers have noted that while these AI offerings are priced at $1 for now, the arrangements are temporary. Agencies relying on these tools may face higher costs when the promotional terms expire, raising questions about future procurement and vendor lock-in.

GSA eyes fast-tracked approvals for AI tools

• Following these offers from OpenAI and Anthropic, GSA is exploring ways to expedite security reviews for AI tools across federal agencies.

  • While FedRAMP clearance is a potentially significant barrier, GSA is now consulting with the Federal CIO Council to establish a prioritization pathway for AI vendors newly added to the government’s purchasing schedule.

  • In parallel, GSA is reportedly working with OpenAI on a temporary authorization-to-operate (ATO) process for ChatGPT Enterprise. This could allow some agencies to adopt the tool based on GSA’s internal review, pending full FedRAMP clearance.

    • The agency has signaled its preference for short initial terms while agencies build familiarity with generative AI.

    • Broader FedRAMP reforms are also underway, aimed at streamlining the approval process for emerging technologies.

Subscribe now

Data center industry urges Treasury to preserve clean energy subsidy rules

• Google, Amazon, Microsoft and other data center operators have urged Treasury Secretary Scott Bessent to retain current clean energy subsidy rules, warning that stricter requirements could stall new energy projects at a time of rising electricity demand from AI infrastructure.

  • The letter, dated August 4 and organized by the Data Center Coalition, comes in response to a July executive order from President Trump directing the Treasury to tighten eligibility for wind and solar tax credits.

  • “Any regulatory friction that slows down deployment of new generation today directly impacts our ability to meet AI-era electricity demands tomorrow,” the coalition wrote.

    • Industry analysts estimate that up to 60 gigawatts of planned solar capacity could be lost by 2030 under the new rules.

    • The data center sector contributed $3.5 trillion to US GDP between 2017 and 2023, according to the coalition, and supports over 600,000 jobs.

    • The Treasury Department is expected to release final guidance by August 18.

    

NSF and NVIDIA invest in open AI models for science

• The National Science Foundation and NVIDIA are jointly investing over $150 million to support the development of open-source AI models for scientific research.

  • The project, led by the Allen Institute for AI (Ai2), will build multimodal models trained on scientific literature and data.

  • The initiative aligns with the Trump administration’s AI Action Plan, which prioritizes open-weight models and public-private collaboration. NSF is contributing $75 million; NVIDIA will provide $77 million.

  • NSF said the project aims to accelerate scientific discovery by making advanced AI tools freely available to researchers.

Subscribe now

Press Clips

• Transistor Radio: H20, GPT5, DUI Lawyers, Ajinomoto (ChinaTalk) 🔉

• Meta’s Guide to Governance: Bulldoze, Lobby, Repeat (AI Governance, Ethics and Leadership) ✍

• China is killing the US on energy. Does that mean they’ll win AGI? (Casey Handmer, Dwarkesh Podcast) 🔉

• Out of Thin Air (Dean Ball, Hyperdimensional) ✍

• AI & Jobs: Enter Populism? (Anton Leicht, Threading the Needle) ✍

• Avoiding a data center electricity price apocalypse (Matthew Yglesias, Slow Boring) ✍

• Episode 13: Reporting in Taiwan and China (Special Competitive Studies Project, Strait Forward Podcast) 🔉

Delivered to your inbox every Monday, Last Week in AI Policy is a rundown of the previous week’s happenings in AI governance in America. News, articles, opinion pieces, and more, to bring you up to speed for the week ahead.

Policy

U.S. is exploring location tracking for chips, says Michael Kratsios

• Science and Technology Policy Advisor and architect of the US AI Action Plan Michael Kratsios described the potential implementation of “software or physical changes you could make to the chips themselves to do better location-tracking.”

  • Physical mechanisms for location tracking and other security measures are often described as hardware-enabled mechanisms and a form of on-chip governance.

  • They could allow for better execution of export controls on chips, which in recent months have struggled as a result of extensive black markets and backdoors into restricted geographies.

NVIDIA chip shipments delayed by export approvals

• Under Commerce Secretary Howard Lutnick, the Department is facing a backlog of thousands of export license applications.

• Among the hardest hit companies is NVIDIA, who are currently trying to meet enormous demand from China.

  Thanks for reading! Subscribe for free to receive new posts and support our work.

TSMC’s stock soars as it benefits from tariff exemption

• Taiwan Semiconductor Manufacturing Company (TSMC), the world’s largest chip manufacturer, will not have to pay Trump’s 100% tariff on chips.

• “TSMC is exempted from the chip tariffs because it has set up plants in the US,” said Taiwanese minister Liu Chin-ching.

  • TSMC’s shares rose to record high on Thursday.

Two Chinese nationals arrested for alleged illegal exports of chips

• They were charged with violating the Export Control Reform Act by exporting tens of millions of dollars worth of chips to China.

  • Through a California-based company, ALX Solutions Inc., the pair allegedly made at least 20 shipments from the U.S. to China, via freight-forwarding companies in Malaysia and Singapore.

• They face felony charges, carrying a maximum sentence of up to 20 years in federal prison.

NSF invests $100M to expand AI Research Institutes

• Following the release of the AI Action Plan, the National Science Foundation announced a $100 million public-private investment to expand its National AI Research Institutes.

• The funding will support five institutes and a central hub focused on generative AI, STEM education, drug discovery, and advanced materials. New projects include research into safer AI assistants, clinical imaging tools, and molecule design.

GAO details federal use of generative AI

• A new GAO report finds federal agencies are rapidly adopting generative AI, with reported use cases rising from 32 in 2023 to 282 in 2024.

• Twelve agencies were reviewed; eleven reported generative AI activity in 2024, with NASA as the sole exception. Nearly one-third of use cases remain in early development stages.

SEC launches AI Task Force to drive innovation and efficiency

• The Securities and Exchange Commission has launched an agency-wide task force to advance the responsible use of AI. Valerie Szczepanik, newly appointed as the SEC’s Chief AI Officer, will lead the effort.

  • The task force will centralize AI initiatives across SEC divisions and offices, support cross-agency collaboration, and guide the adoption of AI tools. It will also focus on aligning innovation with governance and accelerating mission-aligned AI deployments.

Subscribe now

GSA eyes expansion of internal AI tool to other agencies

• The General Services Administration is exploring a government-wide rollout of its internal generative AI platform, GSAi.

• Chief AI Officer Zach Whitman described the effort as the “next iteration” of GSAi, allowing other agencies to adopt it in isolated, tenant-based environments tailored to their needs.

  • GSAi, launched internally in March, draws on models from OpenAI, Anthropic, and Google, and is used by nearly half of GSA staff daily.

    • Whitman stressed a “vendor-agnostic” approach and said GSA’s AI Safety Team is evaluating tools like xAI’s Grok 3 and 4. He said expanding GSAi could cut costs and reduce duplicated agency contracts: “Everyone’s building the exact same thing.”

Presidential AI Challenge aims to build youth skills through community-based projects

• As part of the Trump administration’s efforts to bolster AI education, the White House has launched the Presidential AI Challenge to encourage students and educators to apply AI tools to local and regional problems.

  • The initiative seeks to expand K–12 engagement with AI by fostering early exposure to responsible use and real-world application.

  • The program aims to build national capacity in AI literacy and prepare the next generation for participation in an AI-assisted workforce.

Subscribe now

Press Clips

• What Happens When AI Schemes Against Us (Garrison Lovely, Obsolete) ✍

• AI Education: Understanding the Hype (Lily Ottinger and Irene Zhang, ChinaTalk) ✍

• How not to lose your job to AI (Benjamin Todd, 80,000 Hours Podcast) 🔉

• Episode 83: Horacio Rozanski on Delivering Tech Solutions Rapidly to the Warfighters (Special Competitive Studies Project) 🔉

• Human-level AI is not inevitable. We have the power to change course (Garrison Lovely, The Guardian) ✍

• Episode 12 - Cyber Statecraft (Special Competitive Studies Project) 🔉

• Robotics Levels of Autonomy (SemiAnalysis) ✍

• GPT-5 and the arc of progress (Nathan Lambert, Interconnects) ✍

• AI disagreements (Brian Merchant, Blood in the Machine) ✍

• A discussion with Tyler Cowen (AI Policy Perspectives) 🔉

• Keeping AI agents under control doesn't seem very hard (Timothy B. Lee, Understanding AI) ✍

• AI Is Not A Product (Anton Leicht, Threading the Needle) ✍

• Quantifying the algorithmic improvement from reasoning models (Anson Ho and Arden Berg, Epoch AI) ✍

• Human judgment and artificial intelligence (Nicklas Berild Lundblad, Unpredictable Patterns) ✍

• Opportunities to Strengthen U.S. Biosecurity from AI-Enabled Bioterrorism: What Policymakers Should Know (Georgia Adamson and Gregory Allen, CSIS)