AI Policy · Daily

President Trump signed a voluntary AI cybersecurity executive order giving the federal government up to 30 days to review frontier models before release, with a Treasury-led AI cybersecurity clearinghouse. Sen. Kirsten Gillibrand introduced legislation restricting Pentagon AI use in nuclear launch decisions, domestic surveillance and weapons development, teeing up an NDAA fight as the House Armed Services Committee marks up Thursday. The UK Competition and Markets Authority ordered Google to let publishers opt out of having content used in AI Overviews and AI Mode, with nine months to comply and biannual reporting for the first year. Sen. Bernie Sanders previewed an American AI Sovereign Wealth Fund Act that would take a 50% public ownership stake in the largest U.S. AI firms through a one-time stock tax, with proceeds directed to cash payments and social programs.

I.Top Stories

President Trump signs voluntary AI cybersecurity executive order with 30-day prerelease window and Treasury-led clearinghouse

President Donald Trump signed an executive order Tuesday asking AI companies to give the federal government up to 30 days to review frontier models before public release, the New York Times reported. The order disclaims any mandatory licensing, preclearance or permitting requirement. It directs the Treasury secretary to form an AI cybersecurity clearinghouse, in consultation with the National Cyber Director, the NSA and CISA, to discover and validate software vulnerabilities. The 30-day window was shortened from the 90 days in the version Trump declined to sign on May 21, and White House AI advisor David Sacks publicly called the new timeline a "game changer" that lets labs comply without release delays. Microsoft President Brad Smith, OpenAI Chief Global Affairs Officer Chris Lehane and Google President of Global Affairs Kent Walker each praised the order as a step balancing AI safety and innovation.

Read at NYT ↗ • Read at CNBC ↗ • Read at The White House ↗

AI Policy Network welcomes EO national-security focus and asks Congress to legislate durable measures beyond cybersecurity

The AI Policy Network said Tuesday it welcomed the focus on AI and national security in President Trump's executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security." AIPN President Mark Beall said, "AI companies have created superhuman cyberweapons, but the same capability jumps will soon extend to other domains, like novel pathogen development and missile design... Congress should build on this EO and create a permanent framework, with the ability to not only evaluate but also direct action in response." The organization urged Congress to enact durable security measures extending beyond cybersecurity.

Read at AIPN ↗

Sen. Gillibrand introduces Secure and Accountable Military AI Act limiting Pentagon AI in nuclear launch, surveillance and autonomous weapons

Sen. Kirsten Gillibrand (D-N.Y.) introduced legislation Tuesday that would limit the use of AI for launching nuclear weapons, surveilling Americans and developing or deploying autonomous weapons, and require human intervention plus senior-official approval for "high-consequence actions," The Hill reported. She plans to offer elements as an amendment to the Senate NDAA, which the Senate is expected to begin marking up next week. The House Armed Services Committee separately debates its version Thursday, June 4.

Read at The Hill ↗

UK CMA orders Google to give publishers opt-out controls over AI Search summaries with nine months to implement

The UK Competition and Markets Authority imposed binding conduct requirements on Google's search services Wednesday, allowing publishers to opt out of having their content used in AI Overviews and AI Mode and for fine-tuning of AI models, The Guardian reported. The CMA acted under the new digital markets competition regime after designating Google with strategic market status in general search, where Google accounts for more than 90% of UK queries. CMA Chief Executive Sarah Cardell said publishers, including news organizations, need appropriate bargaining power over how their content is used. Google has nine months to implement all the changes and must submit compliance reports every six months for the first year, per Reuters.

Read at The Guardian ↗ • Read at Investing.com ↗ • Read at GOV.UK ↗

Sen. Sanders previews American AI Sovereign Wealth Fund Act with 50% public stake in largest U.S. AI firms

Sen. Bernie Sanders (I-Vt.) said Monday he will introduce the American AI Sovereign Wealth Fund Act in the coming weeks, a bill that would give the public a 50% ownership stake in the largest American AI companies, The Hill reported. Sanders described the proposal in a nearly seven-minute video message and a New York Times op-ed, framing the stake as a one-time 50% tax paid in stock rather than cash, modeled on Norway's Government Pension Fund Global and the Alaska Permanent Fund, per Tom's Hardware. The federal government would hold voting shares and equal board representation at OpenAI, Anthropic and xAI under the proposal, with proceeds used for direct payments to Americans and for funding health care, education and housing.

Read at The Hill ↗ • Read at Tom's Hardware ↗ • Read at Sen. Sanders ↗

OpenAI's Altman makes Washington swing for White House and bipartisan Hill meetings days after Trump's AI executive order

OpenAI CEO Sam Altman meets Wednesday with high-level White House officials and bipartisan members of Congress, Politico Pro reported, citing a person familiar with his plans. The Washington, D.C. visit, planned in advance, lands days after President Trump signed a modified executive order Tuesday creating a voluntary 30-day government review for advanced models before public release, which OpenAI had spent weeks discussing with the administration. Altman is expected to discuss safety policy, where the models go next, and cyber capabilities, arriving as OpenAI's latest model, GPT-5.5-Cyber, rolls out as a competitor to Anthropic's Mythos.

Read at Politico Pro ↗

II.China Watch

Tencent prepares to file WeChat AI agent with Chinese regulators this month

Tencent is testing a prototype AI agent embedded inside WeChat and plans to begin the regulatory approval process required for a public rollout as early as this month, TechNode reported, citing people familiar with the matter. The agent is designed to complete tasks across WeChat's mini-program ecosystem on a user's behalf, accessed by swiping right from the chat home screen. A full launch is targeted for Q3 2026, contingent on clearance from the Cyberspace Administration of China under Beijing's generative AI services rules. WeChat has roughly 1.4 billion users.

Read at TechNode ↗

YMTC's global flash memory share jumps to 13% as AI demand drives Chinese catch-up

Yangtze Memory Technologies' share of global NAND flash revenue rose to 13% in the first quarter of 2026, up from 8% a year earlier and now level with U.S. memory makers Micron and SanDisk, Caixin reported, citing research firm Counterpoint. The global NAND market hit a record $46 billion in the quarter, more than triple the year-earlier figure, driven by AI infrastructure rollouts and enterprise solid-state drive demand. A successful initial public offering and capacity expansion could make the Entity List company the world's third-largest flash maker.

Read at Caixin ↗

MiniMax releases open-weight frontier coding model and undercuts U.S. labs on price

Shanghai-based MiniMax launched M3 on June 1, a large language model with a million-token context window and frontier coding scores. A QbitAI hands-on test put its benchmark performance ahead of OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro and approaching Anthropic's Claude Opus 4.7 at roughly one-tenth of leading proprietary pricing. MiniMax plans to release the model to Hugging Face within roughly 10 days of launch and supports private cluster deployment, though it has stopped short of a full open-source license. The launch adds to a wave of Chinese open-weight releases that have closed the capability gap with frontier U.S. systems.

Read at QbitAI ↗

III.Federal Policy Tracker

USTR opens public comment window on creating U.S.-China Board of Trade, with nonsensitive product tariff cuts in scope

The Office of the U.S. Trade Representative on Tuesday opened a public comment process on the U.S.-China Board of Trade that President Trump and Chinese President Xi Jinping agreed to at their mid-May summit in Beijing, with comments due July 10, per Reuters. USTR is seeking input on which Chinese products and U.S. exports could be considered nonsensitive and qualify for tariff modifications under the new mechanism, with both sides describing about $30 billion worth of goods on each side as initially in scope. U.S. Trade Representative Jamieson Greer said the administration would work with stakeholders to identify nonsensitive goods trade that can deliver results for American farmers, ranchers, fishermen, small businesses, manufacturers and workers.

Read at Bloomberg ↗ • Read at StreetInsider ↗

IV.Industry & Market Watch

Anthropic expands Project Glasswing and Mythos access to about 150 new organizations across 15-plus countries including NATO

Anthropic is expanding Project Glasswing, its security vulnerability program, and access to its Mythos model to about 150 new organizations across 15-plus countries in sectors covering power, water, healthcare, communications and hardware, TechCrunch reported. Named allied participants include Okta, Samsung, SK Hynix, SK Telecom and NATO. Anthropic estimates that a successful cyberattack on partner codebases could affect more than 100 million people, and described Mythos as capable of identifying thousands of zero-day vulnerabilities. The April cohort included 50 partners, among them the U.S. government.

Read at TechCrunch ↗ • Read at Anthropic ↗

U.S. data center buildout slipping behind schedule even as Alphabet raises $80 billion for AI infrastructure

America's data center buildout is falling behind schedule, with power supply, transformer shortages and permitting fights cited as the binding constraints, the Wall Street Journal reported. A JP Morgan analysis found more than 60% of capacity planned for 2027 isn't yet under construction. Alphabet is raising a fresh $80 billion — $10 billion from Berkshire Hathaway plus ~$70 billion by other means — to fund the build-out, though its workaround for the power bottleneck is owning generation (its $4.75 billion Intersect acquisition) and load-shifting rather than the cash itself. Markets received the raise warily: Alphabet shares fell 3.9%, a $340 billion three-day loss.

Read at WSJ ↗

Ex-DOGE staffers launch AI driven cost-cutting venture Special, backed by Musk allies and Andreessen Horowitz

Two former Department of Government Efficiency staffers, Nate Cavanaugh and Justin Fox, unveiled a new venture called Special on Tuesday to buy companies and cut waste by implementing AI, applying the DOGE cost-cutting playbook to the private sector, Bloomberg reported. The firm raised funding from Musk allies including Valor Equity Partners founder Antonio Gracias, former xAI Chief Financial Officer Anthony Armstrong, and Steve Davis, who served as Musk's de facto second-in-command at DOGE. Andreessen Horowitz led an undisclosed investment in the firm.

Read at Bloomberg ↗

Google publishes five water commitments including a 2030 net positive replenishment target for data centers

Google laid out five commitments on water use in a blog post Wednesday, including a goal to replenish more water than it uses at its data centers by 2030, The Verge reported. Google also said it will invest in local water infrastructure, identify alternative water sources to power its facilities and be transparent about its water use overall. The post arrives amid pending state public utility commission and water board decisions on AI data center permitting.

Read at The Verge ↗

Amazon faces class action lawsuit in Seattle over Ring "Familiar Faces" facial recognition feature

A Virginia resident named Charles Sigwalt filed a class action lawsuit against Amazon in federal court in Seattle on Monday, alleging that Ring's optional "Familiar Faces" feature stores images of passersby without consent and violates privacy rights, TechCrunch reported. The complaint focuses on biometric image retention by the Ring doorbell camera feature, claiming Familiar Faces collects and stores facial data without meaningful opt-in. Sigwalt is seeking certification of a nationwide class of people whose facial recognition data Familiar Faces allegedly collected, along with a Virginia subclass, and at least $5 million in damages.

Read at TechCrunch ↗

V.Global & Geopolitics

UK Commons Tech Committee urges activation of 2027 break clause to end Palantir's NHS data role

The UK House of Commons Science, Innovation and Technology Committee published a report Wednesday calling on the government to trigger a 2027 break clause in Palantir's £330 million ($444 million) NHS Federated Data Platform contract and to seek a UK provider or in-house alternative, the Financial Times reported. Committee Chair Dame Chi Onwurah said reliance on a small number of U.S.-based providers represents a clear vulnerability, and that vendor lock-in produces worse, more expensive services over time. The 11-member bipartisan committee called Palantir's UK public services role "an unacceptable point of weakness" and pointed to past public comments by Palantir cofounder Peter Thiel and CEO Alex Karp as evidence of a mismatch with UK values, per WIRED. The committee also asked the government to disclose additional details of Palantir's UK military contracts.

Read at FT ↗ • Read at WIRED ↗ • Read at Sky News ↗ • Read at Reuters ↗

EU €20 billion AI gigafactory plan stumbles as bidding slips to July and interested consortia narrow to 10

The EU's €20 billion ($23.3 billion) plan to build five AI gigafactory data centers is faltering before bids are submitted, Bloomberg reported. The call for proposals has been pushed from May to July, and funding clarity is available for only two of the five sites before the next EU budget cycle in 2028. Interest has narrowed from about 70 companies in the initial sounding round to roughly 10 expected bidders, with the Schwarz Group and at least one other consortium reconsidering their participation, per The Next Web. SoftBank's separate €75 billion ($83 billion) France data center plan alone exceeds three times the total EU program, and Mistral AI CEO Arthur Mensch criticized the national framing of the EU initiative. The European Commission's €4.1 billion ($4.5 billion) in direct subsidies are split across the five planned country sites, with member states matching and private investors covering the balance.

Read at Bloomberg ↗ • Read at The Next Web ↗