AI Policy · Daily

The Trump administration is pressing Meta to submit frontier AI models for voluntary federal vetting, leaving it the lone holdout among major U.S. developers under the June 2 review framework. The NSA lost access to Anthropic's Mythos model after a June 12 export directive barred foreign-national use, forcing the company to disable its top systems worldwide despite their demonstrated ability to find classified-system vulnerabilities in hours — and an AI startup has since filed the first court challenge to the directive. RAISE Act author Alex Bores lost his NY-12 Democratic primary to establishment favorite Micah Lasher, capping a Manhattan race that AI-industry spending on both sides, more than $20 million, had turned into a national proxy war over AI regulation. The Energy Department unveiled a $17.5 billion loan program for 10 new nuclear reactors to power AI data centers, with up to five conditional loans backing Westinghouse AP1000 builds.

I.Top Stories

Trump administration presses Meta to volunteer AI models for federal review

The Trump administration is pressing Meta to submit its frontier AI models for voluntary federal review, in a confidential request made by email and described to the New York Times by four people familiar with the matter. Meta is the only major U.S. developer that has not signed an agreement to share models with the federal government; OpenAI, Anthropic, Google DeepMind, Microsoft and xAI already participate through the Center for AI Standards and Innovation. Meta, which launched its Muse Spark model in April, told Reuters it expects to sign after working through details. The push implements the June 2 executive order, which established a voluntary framework letting officials review "covered frontier models" for up to 30 days before public release.

Read at NYT ↗ • Read at The Straits Times ↗ • Read at Anadolu ↗

NSA loses access to Anthropic's Mythos under Trump export directive

The National Security Agency lost access to Anthropic's Mythos model after the administration's June 12 export directive barred foreign-national use of the company's most advanced systems, with Anthropic disabling both Mythos 5 and Fable 5 worldwide to comply, per the New York Times. The same Mythos model had identified vulnerabilities in classified U.S. government computer systems "not in weeks but in hours" during testing under Anthropic's Project Glasswing initiative, Sen. Mark Warner told a June 11 Senate Banking hearing. Warner attributed the assessment to NSA and Cyber Command chief Gen. Joshua Rudd. Anthropic called the directive a "misunderstanding," saying the government had not specified the national security concern but that Anthropic had complied to keep faith with the export controls framework. The NSA declined to comment.

Read at AP ↗ • Read at The Washington Post ↗ • Read at NYT ↗

RAISE Act author Alex Bores loses NY-12 primary that AI spending turned into a national proxy war

Alex Bores lost Tuesday's NY-12 Democratic primary in Manhattan to Micah Lasher, the party establishment's favorite, the Financial Times reported. Heavy AI-industry spending on both sides had turned the underdog's bid into a national proxy war over AI regulation, with strategists crediting the onslaught for sharply raising his profile. The OpenAI-tied Leading the Future PAC spent roughly $7.6 million against him through a subsidiary funded by Greg Brockman, Marc Andreessen and Joe Lonsdale, while Anthropic-backed groups spent more than $10 million in his support and crypto billionaire Chris Larsen pledged another $3.5 million. Bores wrote New York's RAISE Act, which requires frontier AI developers to report safeguards against catastrophic risks that could injure more than 50 people — the kind of state rule the Trump administration's proposed federal preemption would nullify.

Read at FT ↗ • Read at AP ↗ • Read at The Hill ↗

DOE unveils $17.5B loan program for 10 new nuclear reactors to power AI data centers

The Trump administration will offer $17.5 billion in conditional loans to support 10 new large-scale nuclear reactors designed to meet surging power demand from AI data centers, Energy Secretary Chris Wright said Tuesday, per The Hill. DOE plans to issue up to five conditional loans of roughly $3.5 billion each, with each award supporting two reactors using the Westinghouse AP1000 design built at Plant Vogtle in Georgia. Seven utilities and energy companies have signed letters of intent naming sites, and DOE will pick five to host construction expected to begin by 2030 and operations by the mid-2030s. The federal money will pay for long-lead nuclear components rather than construction itself, and slots into Trump's stated goal of quadrupling domestic nuclear capacity within 25 years.

Read at AP ↗ • Read at The Washington Post ↗ • Read at The Hill ↗

AI startup Legion sues U.S. over loss of Fable 5 access

An AI startup that uses Anthropic's models sued the federal government Tuesday over the June 12 export directive that forced Anthropic to take Fable 5 and Mythos 5 offline worldwide, Bloomberg reported. The directive, framed as an export control against foreign-national use, came 10 days after Trump's June 2 executive order establishing a voluntary review framework for advanced AI models. Anthropic disputed the order, saying the government had not specified the national security concern. It disabled both models for all customers to comply. The lawsuit is the first court challenge to the export directive.

Read at Bloomberg ↗ • Read at AP ↗ • Read at TechCrunch ↗

II.China Watch

China retakes the top supercomputer ranking with a fully domestic, GPU-free build

China returned to No. 1 on the global TOP500 supercomputer ranking with Lingsheng, a domestic system that hit 2.19 exaflops of sustained performance at the ISC 2026 conference in Hamburg on June 23, per Pandaily. The result ended El Capitan's run at the top and marked China's first time leading the rankings since Sunway TaihuLight in 2017. Lingsheng pairs domestic ARM-based processors with China's first homegrown high-bandwidth memory and the Kylin operating system, with no Nvidia or AMD accelerators in the stack. Lu Yutong, director of the National Supercomputing Center in Shenzhen, said the design pioneers an Online Acceleration architecture that breaks the conventional CPU-GPU split.

Read at Pandaily ↗

Bessent says China catching up on AI is a bigger U.S. risk than safety or jobs

Treasury Secretary Scott Bessent told the Economic Club of New York on Tuesday that China overtaking the United States on AI is the biggest risk the technology poses, ranking it above safety or job-loss concerns, per the South China Morning Post. Bessent described himself as one of the administration's point people on AI policy and its point person on the U.S.-China economic relationship. He said Beijing's willingness to enter formal AI talks with Washington is itself evidence that the U.S. holds the lead. The two governments agreed to begin those talks after the Trump-Xi summit in Beijing last month.

Read at SCMP ↗

Lawyers describe a "China Initiative 2.0" intensifying U.S. visa and investigation pressure on Chinese researchers

Leading immigration lawyers and activists describe the current U.S. crackdown on Chinese scientists and academics as more aggressive than the 2018 "China Initiative," which was scrapped in 2022, the South China Morning Post reported. Robert Fisher, a former Assistant U.S. Attorney now at Nixon Peabody, said federal and state investigations involving China-linked researchers are seeing a "large uptick" and that "we are clearly in an era of China Initiative 2.0," though most have not yet produced charges. In recent months the FBI has arrested researchers tied to universities, accusing them of smuggling biological materials and concealing ties to Chinese state institutions; Indiana University scholar Youhuang Xiang pleaded guilty in April to smuggling E. coli samples and was ordered deported. Lawyers said scrutiny is concentrated in fields central to U.S.-China competition, including AI, robotics, semiconductors and biotech, with a bipartisan push in Congress now producing bills that mirror the original initiative.

Read at SCMP ↗

Alibaba sues Pentagon to be removed from Chinese military companies list

Alibaba sued the U.S. Department of Defense on Tuesday in San Jose federal court seeking removal from the Section 1260H list of Chinese military companies, weeks after the Pentagon expanded the blacklist to 188 entities on June 8, per the Financial Times. The complaint argues the designation has "no basis in fact or law," violates due process and impugns Alibaba's commercial reputation. Alibaba said it is governed by an independent board with no military ties and that its products are built for retail, logistics and enterprise IT. The Pentagon accused Alibaba of being a "military-civil fusion contributor to the Chinese defense industrial base" through an affiliation with China's Ministry of Industry and Information Technology and indirect ties to state asset regulator SASAC. Recent additions to the list also include Baidu, BYD, NIO and WuXi AppTec, which filed a similar challenge June 11.

Read at NDTV ↗ • Read at The Straits Times ↗ • Read at FT ↗

Nvidia's banned AI chips more than double on China's black market over six months

Banned Nvidia AI chips have more than doubled in price on China's black market over the past six months as U.S. enforcement actions cut illegal export channels, the Financial Times reported. The Nvidia DGX B300 server, which contains eight Blackwell GPUs and lists for about $400,000 in the U.S., has climbed to more than 8 million yuan ($1.1 million) from 4 million yuan ($560,000), per traders surveyed by the FT. The RTX 6000 Pro workstation chip, used by startups deploying large language models, rose to as much as 130,000 yuan ($18,000) from about 50,000 yuan ($7,000) at the start of the year. In March, a Supermicro co-founder, a Taiwan-based employee and a contractor were charged with smuggling $2.5 billion in Nvidia AI servers to Chinese customers, a U.S. enforcement action that traders cited as one driver of the price spike.

Read at Investing.com ↗ • Read at FT ↗

III.Policy Tracker

Sens. Curtis and Schiff unveil bipartisan SAFE KIDS Act regulating AI chatbots for minors

Sens. John Curtis (R-Utah) and Adam Schiff (D-Calif.) introduced the Safeguarding AI Features to Ensure Kids' Informed Digital Safety Act on Tuesday and asked that it be considered in an upcoming markup, per Politico. The SAFE KIDS Act would require chatbot providers to conduct ongoing risk assessments before making systems available to children, bar advertising targeted at minors, prohibit bots from generating sexual deepfakes or mimicking human emotions, and require providers to notify parents when a child shows signs of suicidal ideation or self-harm. It would also bar the sale of children's personal data without parental consent and mandate annual independent child safety audits made public. "Parents deserve confidence that AI tools are not exposing their children to harmful content, fostering unhealthy emotional dependence, or exploiting their personal information," Curtis said in a statement.

Read at Bloomberg Law ↗ • Read at KSL.com ↗ • Read at Politico ↗

House passes SBA AI Utilization Act mandating annual agency AI reporting

The House passed the SBA Artificial Intelligence Utilization Act (H.R. 8881), which would require the Small Business Administration to deliver an annual report to Congress on its use of the technology, FedScoop reported. The measure responds to years of SBA noncompliance with federal AI inventory reporting requirements that apply to executive agencies. The bill now heads to the Senate.

Read at FedScoop ↗ • Read at govinfo.gov ↗

DOE launches Quantum Genesis initiative under new quantum executive orders

The Department of Energy launched its Quantum Genesis initiative Tuesday to coordinate federal labs toward fault-tolerant quantum computing by 2028, FedScoop reported. The program has three priorities: a competition to develop quantum systems, a supercomputing facility for quantum engineers, and targeted research to advance specific scientific use cases. Quantum Genesis pairs with the Genesis Mission's high-performance computing platform to give academic and corporate developers access to federal quantum infrastructure.

Read at FedScoop ↗

Nevada's narrow 2025 AI laws stand out as states regulate despite Trump preemption push

Nevada adopted targeted AI measures in its 2025 legislative session covering education, mental and behavioral health, and emergency planning, taking a narrower approach than other states even as the Trump administration warns against state-level AI regulation, per The Nevada Independent. The laws, AB406 and AB325, focus on specific harms rather than the broad developer-accountability frameworks pursued in New York's RAISE Act or California; a more comprehensive Nevada bill, SB199, failed. The measures feature in a new AP roundup showing states forging ahead while federal AI regulation remains stalled.

Read at The Nevada Independent ↗

IV.Industry & Market Watch

OpenAI and Broadcom unveil Jalapeño, OpenAI's first in-house AI chip

OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom-designed AI chip, an inference accelerator OpenAI is already testing in its labs and plans to put into commercial use late this year, per Axios. OpenAI did the core chip design while Broadcom contributed networking and connectivity, and OpenAI said Jalapeño delivers substantially better performance per watt than off-the-shelf options. OpenAI said it took the chip from design to manufacturing tape-out in nine months, using its own models to speed parts of the work. The move pushes OpenAI deeper into custom silicon as it works to secure compute and cut its near-total reliance on Nvidia, which it said remains a key partner for training.

Read at OpenAI ↗ • Read at Bloomberg ↗ • Read at Axios ↗

Tesla, Sunrun and Renew Home pitch a residential virtual power plant for AI data centers

Tesla, Sunrun and Renew Home unveiled a plan to aggregate residential solar panels, batteries and smart thermostats into a virtual power plant to help meet surging power demand from AI data centers, the New York Times reported. By dispatching stored power and trimming demand during peaks, the companies say the network could free up enough capacity to serve 17 large data centers and defer some of the new power plants and transmission lines that otherwise cost billions and take years to build. The proposal dovetails with FERC's June 18 large-load interconnection orders, which aim to speed data centers onto the grid, in part by letting those that curb usage during peak strain connect faster.

Read at NYT ↗ • Read at FERC ↗

Agility Robotics to go public in $2.5B SPAC merger

Humanoid robot maker Agility Robotics announced it will go public through a merger with Churchill Capital Corp XI in a transaction valuing the company at about $2.5 billion, the Wall Street Journal reported. The combined company will trade as AGLT and is expected to raise more than $600 million in gross proceeds, including roughly $420 million from Churchill's trust account and more than $200 million from a private investment in public equity led by Taiwan-based Foxconn. Agility's flagship Digit humanoid is in use at Amazon, logistics firm GXO, auto parts supplier Schaeffler and Toyota Motor Manufacturing Canada, with the company backed by Amazon, Nvidia and SoftBank.

Read at WSJ ↗ • Read at Investing.com ↗

WaPo profiles Amy Kremer's pivot from Jan. 6 organizer to campaigner against AI data centers

The Washington Post profiled Amy Kremer, the longtime Tea Party figure who helped organize the Jan. 6, 2021, Capitol rally. Kremer now chairs Humans First, the conservative group campaigning to rein in AI companies and block data center construction. The profile frames Kremer's pivot to AI policy as the latest extension of a populist movement she helped build through the Tea Party era. Humans First is staging a 13-location nationwide protest day against AI data centers on July 18 across Georgia, California, Texas, Florida and Virginia, previewed by Axios in AIPD's June 22nd edition.

Read at The Washington Post ↗

V.Global & Geopolitics

Dutch Trade Minister lobbies Washington to drop MATCH Act controls on ASML's China sales

Dutch Trade Minister Sjoerd Sjoerdsma met with Commerce Secretary Howard Lutnick and U.S. lawmakers in Washington on June 23 to lobby against the bipartisan MATCH Act, which would expand U.S. controls on ASML's deep ultraviolet immersion (DUV) lithography sales to China, per Bloomberg. China accounted for 19% of ASML's net system sales in the first quarter, down from 36% the prior quarter. "It's exceptional that I'm coming here to broadly outline our concerns to Congress," Sjoerdsma told reporters, criticizing the bill's extraterritorial reach over Dutch trade policy. The Netherlands also signed its Pax Silica accession declaration during the visit, despite the export controls disagreement.

Read at Bloomberg ↗ • Read at The Straits Times ↗ • Read at Devdiscourse ↗

EU joins Pax Silica, the U.S.-led pact to reduce reliance on Chinese AI supply chains

The European Union formally joined Pax Silica, the U.S.-led grouping of allied governments coordinating to break dependencies on Chinese AI supply chains, the Financial Times reported. The Netherlands signed its own Pax Silica accession declaration the same week, joining South Korea and Japan as members; Taiwan endorsed as a non-signatory. Pax Silica is run out of the State Department by Undersecretary for Economic Affairs Jacob Helberg, who coordinates allied export controls, procurement and chip policy.

Read at FT ↗ • Read at Devdiscourse ↗

Australian senators press government on AI copyright and data center moratorium

Independent Senator David Pocock urged the Australian government to prevent tech companies from training AI models on Australian content, with the Labor cabinet weighing copyright rule changes for AI, The Guardian reported. Greens Senator Sarah Hanson-Young separately called for a moratorium on new data center construction in Australia until "we get the regulations right." Hanson-Young warned the country is heading toward an unregulated AI buildout without binding rules on training data or infrastructure. The interventions come from outside Labor's caucus and would require cabinet adoption to take effect.

Read at The Guardian ↗