AI Policy · Daily

The White House is weighing an FDA-style executive order requiring pre-release safety review of frontier AI models, as Vice President JD Vance flagged autonomous exploit risks to critical infrastructure. Xi Jinping cautioned Trump in Beijing that missteps on Taiwan could tip U.S.-China relations toward conflict during their two-hour May 14 meeting. House appropriators included three AI security provisions in the fiscal 2027 Commerce, Justice, Science draft, including a 50% boost for the Center for AI Standards and Innovation (CAISI), NIST chip hardware security grants and $16 million for Bureau of Industry and Security export control data work. Alt5 Sigma, with Eric Trump as a board observer, signed a 90-day memorandum of understanding with Hangzhou chipmaker Nano Labs to explore U.S. AI infrastructure cooperation, despite congressional Republicans previously flagging Nano Labs over alleged Chinese Communist Party ties.

I.AI Policy Today

Vance warns of AI autonomous exploit risk as WSJ details FDA-style pre-release EO deliberations

Vice President JD Vance warned that autonomous AI exploits could threaten critical infrastructure in Wall Street Journal reporting documenting Trump administration deliberations on an executive order requiring pre-release safety review for frontier AI models, modeled on the Food and Drug Administration approval process. The Center for Data Innovation published a same-day rebuttal arguing pre-approval would slow innovation without improving safety. The reporting advances NEC Director Kevin Hassett's May 7 Fox Business framing of an FDA-style pre-release requirement, as reported by The Hill in AIPD's May 7th edition.

Read at WSJ ↗ • Read at Center for Data Innovation ↗

Xi warns Trump that mishandling Taiwan could push U.S.-China ties into conflict at Beijing summit

Chinese President Xi Jinping told President Trump during their May 14 meeting at the Great Hall of the People that mishandling the Taiwan issue could push the U.S.-China relationship into "conflict," Bloomberg reported. Xi described Taiwan as the most important issue in U.S.-China relations and said clashes could put the entire relationship in jeopardy, according to a Xinhua readout carried by Business Standard. Xi also invoked the Thucydides Trap framing in opening remarks, asking whether the two powers could transcend the rising-versus-ruling power rivalry pattern, per the Straits Times. Trump opened the meeting with praise for his counterpart before the closed-door session, which lasted about two hours.

Read at Bloomberg ↗ • Read at Straits Times ↗ • Read at Business Standard ↗

House Appropriations FY27 CJS marked up three AI security provisions

The House Appropriations FY27 Commerce, Justice, Science Subcommittee included a NIST grant program for AI chip hardware security and location verification standards, $16 million for Bureau of Industry and Security export control data infrastructure and a 50% funding increase for the Center for AI Standards and Innovation, the AI Policy Network said. The BIS allocation directs at least $5 million toward commercial datasets, data fusion and data sharing capabilities for export control enforcement. The bill directs NIST, through CAISI, to administer the chip hardware grant program and develop voluntary standards based on its findings. Subcommittee Chairman Hal Rogers and Ranking Member Grace Meng are credited with the AI security provisions.

Read at AIPN ↗

Eric Trump joins Beijing trip as Trump-linked Alt5 Sigma signs MOU with congressionally flagged Nano Labs

Las Vegas fintech firm Alt5 Sigma, which is rebranding as AI Financial Corporation, entered an April 24 memorandum of understanding with Hangzhou chipmaker Nano Labs covering a 90-day evaluation of U.S. AI infrastructure cooperation, the Financial Times reported. The 90-day evaluation covers U.S. data centers, AI native payment systems, cloud platforms for AI agents and other AI infrastructure projects, per The Business Standard. Eric Trump serves as a board observer at Alt5 Sigma and accompanied his father on the May 14-15 Beijing state visit. House and Senate Republicans previously urged the Securities and Exchange Commission to scrutinize Nano Labs as a "high-risk" entity with alleged Chinese Communist Party ties.

Read at FT ↗ • Read at The Business Standard ↗

NAACP seeks injunction against xAI over nearly 50 unpermitted gas turbines in Mississippi

The NAACP, represented by the Southern Environmental Law Center, asked a court for an injunction against Elon Musk's xAI over its operation of nearly 50 natural gas turbines in Southaven, Mississippi, powering its Colossus 2 data center across the state line in Memphis. The state has so far declined to regulate the turbines because they sit on flatbed trailers, qualifying them as "mobile" power plants under a one-year air pollution exemption. The Southern Environmental Law Center argues the turbines violate federal law, which classifies trailer-mounted power plants as stationary and subject to air pollution rules. xAI has been granted permits for 15 of the turbines and is now operating 46, per a local news count.

Read at TechCrunch ↗

DHS plans November autonomous drone reconnaissance exercise along U.S.-Canada border

The Department of Homeland Security, in collaboration with Defense Research and Development Canada, plans a November cross-border exercise using autonomous drones and ground vehicles streaming surveillance video and sensor data over commercial 5G, Wired reported. The DHS call for participants describes the exercise, named ACE-CASPER, as a multiday national emergency response scenario, with feeds relayed to a binational command-and-control center. The document casts vehicle autonomy as secondary to the primary aim of demonstrating "resilient, persistent 5G communications." The tests would be the first joint U.S.-Canada cross-border technology experiment along the shared border in nearly a decade, following the 2011-2017 CAUSE drills.

Read at Wired ↗

Deepfake porn technique profiled five days before FTC's May 19 Take It Down Act deadline

MIT Technology Review documented how nonconsensual deepfake porn and old nonconsensual intimate imagery can resurface via facial recognition matching, profiling a former adult content creator whose 2013 video has been repurposed with other faces grafted on. The reporting traces the practice to generative AI and nudify apps and dates the deepfakes label to a 2017 Reddit username that gave the technique its name. The piece comes five days before the May 19 Take It Down Act compliance deadline that FTC Chairman Andrew Ferguson set for 15 platforms including Amazon, Alphabet, Apple, Meta, Microsoft, Reddit, Snapchat and TikTok, as reported by the FTC in AIPD's May 12th edition.

Read at MIT Technology Review ↗

II.China Watch

Netherlands protests U.S. bill that would further bar ASML from China market

The Dutch government has objected to a proposed U.S. law that would further restrict semiconductor equipment giant ASML from selling to or servicing Chinese customers, the South China Morning Post reported. The Hague lodged a formal protest over the bill's extraterritorial reach, with Dutch officials warning that it could impinge on Dutch sovereignty over export policy. The Multilateral Alignment of Technology Controls on Hardware Act, introduced last month by a bipartisan group of U.S. lawmakers, would block ASML's deep ultraviolet lithography shipments to China and bar servicing of existing customers there.

Read at South China Morning Post ↗

Chinese institutions take roughly 44% of ICLR papers; U.S. share is 32%

Mainland China accounted for about 44% of papers at the International Conference on Learning Representations held in Rio de Janeiro last month, the South China Morning Post reported, citing an analysis of more than 5,000 accepted papers. Tsinghua University, Shanghai Jiao Tong University, Zhejiang University and Peking University took the top four institutional spots globally, with Tsinghua alone producing 332 accepted papers. The U.S. share stood at about 32%, led by Stanford, Carnegie Mellon and MIT, each contributing roughly half as many papers as Tsinghua.

Read at South China Morning Post ↗

WSJ documents accelerating "sea turtle" return of senior Chinese tech talent from Silicon Valley to China

Senior Chinese tech professionals who built careers in the United States, known in Chinese media as "sea turtles," are returning to China in growing numbers and fueling Beijing's effort to take on Silicon Valley, the Wall Street Journal reported. The piece tracks researcher and engineer migration as a parallel competitiveness channel running alongside the chip export control debate. The reporting lands during the May 14-15 Beijing summit, where AI cooperation and chip export carve-outs are on the agenda.

Read at WSJ ↗

III.Federal Policy Tracker

Senate Judiciary subcommittee members defend Copyright Office independence in AI focused oversight hearing

Senators at a Tuesday Senate Judiciary Subcommittee on Intellectual Property hearing defended the Copyright Office's independence amid AI training data litigation and a Trump administration effort to remove Register of Copyrights Shira Perlmutter, IPWatchdog reported. Sen. Thom Tillis said the U.S. should not be in "a race to the bottom" with China on AI policy. The Supreme Court has deferred a decision in Trump v. Perlmutter pending its rulings in Trump v. Slaughter and Trump v. Cook, both heard over the past five months. The House Administration Committee is expected to hold a markup Thursday of H.R. 6028, the Legislative Branch Agencies Clarification Act, which would formally separate the Copyright Office from the Library of Congress.

Read at IPWatchdog ↗

Acting Librarian of Congress requests $5.4 million for centralized enterprise AI platform

Acting Librarian of Congress Robert Newlen asked the Senate Appropriations Legislative Branch Subcommittee on Tuesday for $5.4 million in the fiscal 2027 budget to fund a centralized AI platform for the Library of Congress, FedScoop reported. Newlen said the platform would let the LOC develop and manage AI models in a controlled, secure environment for confidential data, and would enable the Congressional Research Service to clear a bill summary backlog spanning more than 14,000 bills introduced this Congress. He told Sens. Martin Heinrich, D-N.M., and Mike Rounds, R-S.D., co-chairs of the Senate AI Caucus, that the library would "certainly be willing to take the lead" on a platform that could serve the entire Congress. Newlen disclosed he recently spoke with OpenAI CEO Sam Altman about ChatGPT and the LOC's early AI work, without drilling into procurement details.

Read at FedScoop ↗

NRC chair tells senators data reuse will speed but not "rubber stamp" reactor reviews driven by AI data center power demand

Senators questioned the Nuclear Regulatory Commission's fiscal 2027 plan to fast-track reactor licensing by reusing Department of Energy and Department of Defense safety data during a Wednesday budget hearing, FedScoop reported. NRC Chairman Ho Nieh told Sen. Alex Padilla, D-Calif., that the data reuse process would not produce skipped steps or a shortcut through licensing requirements while declining to detail how the agency would validate DOE or DOD data. Nieh said "America urgently needs more energy due to artificial intelligence data centers and industrial growth" and framed the licensing-speed push around AI demand. The White House fiscal 2027 budget proposal would cut NRC overall appropriations by 8.1% and the workforce by 7%, or 196 full-time employees.

Read at FedScoop ↗

OMB federal CIO confirms public release of IT contract data, plans AI review of vendor reporting

Federal Chief Information Officer Greg Barbaccia told FedScoop that the Office of Management and Budget plans to publicly release at least some of the agency IT contract data collected under a March memo, and that OMB will deploy trained AI models against the data to draw correlations. The memo requires certain CIOs to update OMB monthly on contracts they have approved and mandates pricing and usage data collection from vendors. Barbaccia said the current monthly reporting requirements run through September and will be extended, with the agency calibrating the reporting burden across small and large agencies. The disclosure follows critiques that the memo, while citing OPEN Government Data Act standards, did not commit to public dissemination.

Read at FedScoop ↗

UC Berkeley report finds 43 states have established AI governance frameworks

Forty-three states have established some form of AI governance, with wide variation in scope, structure and transparency, per a UC Berkeley School of Information report covered by StateScoop. Eric Hysen, a former Department of Homeland Security chief information officer, argues in the report that AI governance is becoming a core part of state IT operations rather than a standalone policy exercise. The report recommends best practices including AI governance councils, clear human oversight rules and AI governance integration into procurement and cybersecurity workflows. California's Engaged California platform, which gathers resident feedback on AI's workforce and service effects, is cited as a case study alongside Pennsylvania's statewide generative AI pilot.

Read at StateScoop ↗

IV.Capability & Research Watch

Microsoft's MDASH agentic scanner uncovers 16 Windows flaws, including four critical remote code execution bugs

Microsoft detailed a new agentic AI vulnerability discovery system, code-named MDASH, that found 16 previously unknown flaws in Windows networking and authentication components, including four critical remote code execution bugs patched in the May Patch Tuesday release, SiliconANGLE reported. The system orchestrates more than 100 specialized AI agents across frontier and distilled models, spanning the Windows TCP/IP stack, the IKEEXT IPsec service, HTTP.sys, Netlogon, DNS resolution and the Telnet client. MDASH identified all 21 planted vulnerabilities with zero false positives on Microsoft's StorageDrive test driver and recorded 96% recall on five years of Microsoft Security Response Center cases. On the public CyberGym benchmark covering 1,507 real-world vulnerability reproduction tasks, MDASH scored 88.45%, the top result on the leaderboard.

Read at SiliconANGLE ↗

MIT Technology Review documents Google AI chatbot disclosures of real user phone numbers

Google AI chatbots are surfacing real users' phone numbers and other personal contact information, including a case where an Israeli software developer was contacted on WhatsApp after Gemini provided his number in response to a customer service question, MIT Technology Review reported. Privacy-tools firm DeleteMe told the outlet that customer queries about generative AI have increased 400% in the past seven months to a few thousand, with 55% referencing ChatGPT, 20% Gemini and 15% Claude. Experts cited in the piece say the lapses are most likely caused by personally identifiable information appearing in training data, though the precise mechanism producing real phone numbers in AI outputs remains unclear. One PhD candidate at the University of Washington got Gemini to produce her colleague's personal cell phone number while testing the model.

Read at MIT Technology Review ↗

Palo Alto Networks warns AI driven cyberattacks will be the "new norm" within months, with banks named as a vulnerable target

Palo Alto Networks warned that AI driven cyberattacks will become the new norm within months as more sophisticated AI models pressure cybersecurity teams to defend against faster and more autonomous attacks, CNBC reported. Industry experts told EnterpriseAI that Anthropic's Mythos model could accelerate complex cyberattacks against bank legacy systems, with Guardrail Technologies CEO TJ Marlin saying the model can "look across a very complex architecture" to expose previously undiscovered vulnerabilities. The warnings sit alongside Anthropic's Project Glasswing defensive partnership covering Amazon Web Services, Apple, Google, Microsoft and Nvidia.

Read at CNBC ↗ • Read at EnterpriseAI ↗

V.Industry & Market Watch

Texas Hill County approves one-year moratorium on new data center construction

Hill County, Texas commissioners approved a one-year moratorium on new data center and energy storage construction in unincorporated areas in a 3-2 vote Tuesday, citing public safety and public health concerns, Bloomberg reported. The county, roughly 55 miles south of Fort Worth, said the pause appears to be the first issued by a Texas county and follows resident concerns about a proposed 300-acre development by Dallas-based Provident Data Centers in north Hillsboro that would consume large amounts of electricity, per the Texas Tribune. Commissioner Jim Holcomb said developers had called him as late as 10 p.m. the night before the vote asking him to reject the moratorium. County Judge Shane Brassell said the pause will give officials time to study data center effects before projects move forward.

Read at Bloomberg ↗ • Read at Texas Tribune ↗

IDCA pegs data center share of UK and U.S. electricity at 6% as global investment approaches $1 trillion

Data centers consume 6% of electricity in the United Kingdom and the United States, with AI driven energy use up 15% globally over two years and annual global data center investment approaching $1 trillion, or roughly 1% of the global economy, per International Data Center Authority research reported by The Guardian. The figures land amid UK energy shortages and developer reports of multi-year waits for national grid connections. The IDCA called on technology companies to become more transparent about new data center plans to address growing societal and political backlash. The UK government estimated in early 2025 that UK data centers used 2.5% of electricity, well below the new figure.

Read at The Guardian ↗

Cerebras prices upsized IPO at $185 to raise $5.55 billion, valuing the Altman-backed chipmaker at $56 billion

Cerebras Systems priced its upsized IPO of 30 million shares at $185, raising $5.55 billion in the largest U.S. public offering so far in 2026, the Wall Street Journal reported. The deal values the AI chipmaker, backed by OpenAI CEO Sam Altman, at $56.4 billion on a fully diluted basis, per the Reuters wire summary on Investing.com. Pricing came in above the marketed range of $150 to $160 a share, with the offering more than 20 times oversubscribed and shares set to begin trading on Nasdaq under the ticker CBRS on Thursday.

Read at WSJ ↗ • Read at Investing.com ↗

Musk v. Altman testimony reveals Microsoft's early partnership concerns about OpenAI dependence

Senior Microsoft executives testified in the third week of Musk v. Altman in Oakland about early partnership concerns over Microsoft's dependence on OpenAI, CNBC reported. The testimony comes ahead of OpenAI's planned initial public offering and amid Republican state attorneys general pressing the Securities and Exchange Commission to scrutinize OpenAI governance before the listing, as reported by Investing.com in AIPD's May 12th edition. The disclosures add to the record on the antitrust and AGI clause issues shaping the Microsoft-OpenAI commercial relationship.

Read at CNBC ↗