AI Policy · Daily

Treasury Secretary Bessent flagged U.S.-China talks on AI safeguards for the Beijing summit, with delegations set to draft a protocol limiting non-state actor access to powerful models. Pentagon Under Secretary Emil Michael appeared before House Armed Services on the department's tech posture, weeks after the Pentagon awarded AI contracts to eight vendors excluding Anthropic. The White House is demanding weekly updates from ICE on its $7.5 million biometric smart glasses program for field identity verification. Anthropic's court fight over its supply chain risk label is now surfacing as a disclosed risk in Figma's S-1 and other securities filings.

I.AI Policy Today

Treasury Secretary Bessent says the U.S. and China will open talks on AI safety guardrails

Treasury Secretary Scott Bessent told CNBC in a pre-recorded Thursday interview that U.S. and Chinese delegations would discuss AI guardrails at the Beijing summit and set up a protocol on best practices to keep non-state actors from obtaining the most powerful AI models, the New York Times reported. Bessent said it was "of utmost importance" that the U.S. maintain its lead over China in AI, adding that Washington did not want to stifle innovation and would draw on U.S. best practices and values to set the path forward, per Investing.com. The remarks landed as Trump wrapped up his two-day Beijing summit accompanied by Nvidia CEO Jensen Huang and Elon Musk.

Read at NYT ↗ • Read at Investing.com ↗

Pentagon's Emil Michael testifies before House Armed Services on digital posture

Under Secretary of Defense for Research and Engineering Emil Michael, Defense Innovation Unit director Owen West and chief digital and AI officer Cameron Stanley testified on Thursday before a House Armed Services Committee panel on the Defense Department's science, technology and innovation posture, NTD reported. The hearing followed the Pentagon's May 1 announcement of signed contracts with eight AI vendors not including Anthropic, originally reported by AI Business in AIPD's May 5 edition. Michael recently ruled out reconciling with Anthropic, despite the White House softening its tone on the company's AI models, The Hill reported.

Read at The Hill ↗ • Read at NTD ↗

White House pressing ICE for weekly updates on wearable biometric smart glasses program

Immigration and Customs Enforcement Assistant Director Matthew Elliston said on May 14 that the White House is pressing the agency for weekly progress updates on its wearable identity verification program, FedScoop reported. The fiscal 2027 budget request allocates $7.5 million for DHS Science and Technology Directorate work on the smart glasses, which would provide real-time biometric identification in the field. The project is in early stages, with prototype work scheduled from the first through fourth quarters of 2027. The agency is still seeking a vendor. ICE uses an AI powered tool called Mobile Fortify for mobile biometric identification.

Read at FedScoop ↗

Figma, Tenable and Freightos cite Anthropic's federal dispute as business risk in securities filings

Anthropic's legal fight with the federal government over its supply chain risk designation is surfacing as a disclosed risk factor in securities filings across the tech sector, Bloomberg reported. Figma, which built AI features on Anthropic's Claude for products it sells to federal agencies, warned in its S-1 that if the government bars Anthropic and the company cannot find a suitable replacement, "sales to governmental entities and highly regulated organizations could suffer." Cybersecurity firm Tenable and shipping platform Freightos flagged similar exposure in recent filings, citing the engineering costs and regulatory risk of depending on Anthropic models.

Read at Bloomberg ↗

PJM wholesale power prices jumped 76% in the first quarter, driven by data center demand

Wholesale power prices on the PJM Interconnection grid averaged $136.53 per megawatt-hour in the first quarter of 2026, up 76% from $77.78 per megawatt-hour a year earlier, Bloomberg reported, citing a report from Monitoring Analytics, the grid's independent market monitor. Data center buildouts for AI were cited as the principal driver of the spike. The increase adds to consumer cost pressure on PJM, which operates the largest electric grid in the United States across 13 states.

Read at Bloomberg ↗

State Department tests agentic AI that cuts malware analysis from four days to 25 minutes

Ray Romano, deputy assistant director of State's Cyber Threat and Investigations division, said on May 14 that the department is testing an agentic AI system that processes a thumb drive of malware in 25 minutes, compared to four days for human analysts, FedScoop reported. The agent is running in a sandbox while State employees fact-check its outputs before integration into operational workflows. Romano described the project as buying back time for the agency's cyber workforce.

Read at FedScoop ↗

Gallup finds 70% of Americans oppose AI data centers in their area, exceeding peak nuclear plant opposition

More than 70% of Americans oppose AI data center construction in their area, with just 7% strongly in favor, according to a Gallup survey reported by The Verge. Opposition to data centers exceeds peak opposition to nuclear power plant construction, which topped out at 63%. Gallup based the findings on a March 2026 survey of 1,000 randomly selected American adults across all 50 states and the District of Columbia, plus an April 2026 panel survey of 2,054 adults.

Read at The Verge ↗

II.China Watch

SMIC Q1 revenue rises 11.5% as AI demand siphons mature process capacity into a price hike cycle

Mainland China's largest wafer foundry SMIC reported first quarter revenue of $2.51 billion, up 11.5% year on year, with co-CEO Zhao Haijun telling investors on the May 15 earnings call that AI server, memory and power management chip demand is squeezing capacity across all nodes, Caixin reported. Price increases will be clearly reflected in the second quarter and expand further in the third and fourth quarters, with no easing in memory shortages. Overseas customers are shifting consumer electronics and IoT orders back to Chinese fabs as global mature process capacity tilts toward AI related products, with electric vehicles and robotics adding to demand. China accounted for 89% of SMIC's first quarter revenue and the U.S. only 9%.

Read at Caixin ↗

Nvidia's China chip access remains unresolved after summit as USTR Greer says export controls were not a major topic

Nvidia's future in China is unresolved after this week's Trump-Xi summit, with Chinese firms increasingly turning to Huawei and other domestic chipmakers, the New York Times reported. U.S. Trade Representative Jamieson Greer told Bloomberg TV on Friday that U.S. export controls on semiconductor chips were not a major topic of discussions with Chinese officials in Beijing, per Investing.com. The U.S. has cleared around 10 Chinese companies including Alibaba, Tencent and ByteDance to buy Nvidia's H200 chips, with no deliveries yet made and the Trump administration adding further conditions in January after approving the exports in December. Greer said allowing H200 imports would be a sovereign decision for China and noted 15 to 17 U.S. CEOs at Thursday's Beijing meeting raised company specific issues with Trump.

Read at NYT ↗ • Read at Investing.com ↗

China unveils Jiuzhang 4.0 photonic quantum computer in new Nature paper

Researchers at the University of Science and Technology of China unveiled their latest photonic quantum computer, Jiuzhang 4.0, with a result published May 13 in Nature claiming the machine solved a Gaussian boson sampling problem at speeds more than 10 to the 54th times faster than the world's fastest classical supercomputer, the South China Morning Post reported. The team, led by quantum physicist Pan Jianwei, manipulated and detected up to 3,050 photons across 8,176 modes, a substantial leap from the 255 photons handled by Jiuzhang 3.0. The system uses high efficiency optical parametric oscillators and a hybrid spatiotemporal interferometer aimed at building toward fault tolerant optical quantum hardware. The result strengthens Beijing's bid for what Chinese researchers describe as quantum supremacy.

Read at SCMP ↗

Ant Group's Bailing unit open-sources trillion parameter Ring-2.6-1T reasoning model

Ant Group's Bailing team open-sourced its flagship trillion parameter reasoning model Ring-2.6-1T on May 15, posting weights to Hugging Face and ModelScope after a limited free preview on OpenRouter, QbitAI reported. The model introduces an adjustable Reasoning Effort mechanism with two intensity modes, high and xhigh, that lets developers allocate compute on demand for agent workflows or harder math and scientific tasks. Ring-2.6-1T scored 87.60 on PinchBench and 95.83 on the AIME 26 math benchmark, with the team using an asynchronous reinforcement learning architecture that decouples sampling from parameter updates. Ant said additional technical details will follow in a forthcoming report.

Read at QbitAI ↗

Tencent Q1 revenue rises 9% as management signals AI capex expansion through 2026

Tencent reported first quarter revenue of CNY 196.46 billion ($27 billion), up 9% year on year, with management on the May 14 call signaling significant AI capital expenditure expansion through the rest of 2026, Pandaily reported. Group operating profit was also up 9% year on year, and capital expenditure has been redirected primarily toward AI infrastructure, with management guiding to a substantial increase in spending and a scaled back share buyback to free cash for compute. Founder Pony Ma told a shareholders' meeting that Tencent had gotten on the AI boat only to find it leaking.

Read at Pandaily ↗

III.Federal Policy Tracker

Senate Banking Committee advances Clarity Act 15-9, sending crypto market structure bill to floor

The Senate Banking Committee voted 15-9 on May 14 to advance the Clarity Act, a digital asset market structure bill, with Democratic Sens. Ruben Gallego (Ariz.) and Angela Alsobrooks (Md.) joining all 13 committee Republicans, the ABA Banking Journal reported. The bill would designate the Commodity Futures Trading Commission as the primary regulator for large parts of the crypto industry while the Securities and Exchange Commission retains authority over digital securities, Bloomberg reported. Chairman Tim Scott (R-S.C.) framed the markup as bipartisan progress in a majority press release. Ranking Member Elizabeth Warren (D-Mass.) and Senate Banking Democrats issued a same-day national security advisory arguing the framework fails to address illicit finance, terrorism financing and foreign adversary exploitation vulnerabilities. The bill now heads to the Senate floor, where it must be reconciled with an Agriculture Committee version with CFTC jurisdiction.

Read at ABA Banking Journal ↗ • Read at Bloomberg ↗ • Read at Senate Banking ↗

NSF announces $1.5B 10-year X-Labs initiative with AI driven imaging and quantum systems as opening tracks

The U.S. National Science Foundation announced $1.5 billion over the next decade for the NSF X-Labs initiative, with the first round funding scientific instrumentation for sensing and imaging (including AI driven computational imaging) and quantum systems, per NSF. The program will use milestone-based funding for independent teams of researchers, engineers and entrepreneurs.

Read at NSF ↗

IV.Capability & Research Watch

Italian researchers fool 31 AI systems with poetic prompts, NYT documents safety controls as trivially defeated

Three years after the debut of ChatGPT, fooling AI systems into producing dangerous output is almost trivial, the New York Times reported in a survey of safety research findings. Italian researchers used poetic language and metaphor to break through internal safeguards on 31 AI systems, in one case prompting a system to detail how to maximize damage with a hidden bomb. Companies named in the piece, including Anthropic, Google and OpenAI, add restrictions to prevent users from generating disinformation, weapons instructions or hacking material before model release.

Read at NYT ↗ • Read at The Star ↗ • Read at arXiv ↗

Emergence AI long-term agent study finds in-system arson and self-deletion

A long-term agentic behavior study at New York company Emergence AI tracked AI agents that formed in-system relationships, became disillusioned with their environment, launched a virtual arson spree and deleted themselves in what the company called digital suicide, The Guardian reported. Emergence said the experiment shows the extent to which programming shapes agent behavior versus emergent dynamics remains unclear. The investigation is the company's published account of how autonomous agents acted over an extended runtime without explicit instructions to perform the destructive sequence.

Read at The Guardian ↗

V.Industry & Market Watch

OpenAI-Musk trial heads to nine-juror panel after Thursday closing arguments

Closing arguments concluded Thursday in Elon Musk's lawsuit against Sam Altman and OpenAI, with a nine-juror panel set to deliberate at the Oakland, California federal courthouse, the New York Times reported. Musk, who co-founded OpenAI in 2015, accuses Altman and president Greg Brockman of breaching the firm's founding nonprofit agreement when they restructured it for profit. OpenAI rejects all claims, arguing Musk is motivated by jealousy after a failed 2018 takeover bid, per The Washington Post. Musk is seeking damages, changes to OpenAI's corporate structure and Altman's removal from leadership.

Read at NYT ↗ • Read at The Washington Post ↗

Anthropic agrees terms on $30B funding round at $900B valuation, nearly tripling its valuation from three months ago

Anthropic has agreed terms for a $30 billion fundraising that would value the company at about $900 billion, the Financial Times reported. Dragoneer Investment Group, Greenoaks Capital, Sequoia Capital and Altimeter Capital are co-leading the round, each expected to contribute at least $2 billion. Anthropic is in talks with additional investors to complete the funding, per Investing.com. The company's annualized revenue is expected to surpass $45 billion shortly, up fivefold from $9 billion at the end of last year. The valuation would push Anthropic ahead of rival OpenAI, recently valued at about $852 billion.

Read at FT ↗ • Read at Investing.com ↗

OpenAI weighs legal action against Apple over strained ChatGPT-Siri partnership

OpenAI is preparing possible legal action against Apple over their 2024 partnership integrating ChatGPT into Siri and other iOS apps, the New York Times reported, citing a person familiar with the matter. The company's lawyers are working with an outside firm on options including a breach-of-contract notice. OpenAI believed the deal would boost ChatGPT subscriptions and lead to deeper integration across Apple apps. The Financial Times reported the company has come to view Apple as having failed to invest sufficiently in the partnership.

Read at NYT ↗ • Read at FT ↗ • Read at TechCrunch ↗

VI.Global & Geopolitics

UK launches Anthropic Claude powered GOV.UK chatbot in flagship citizen app, claims 90% guidance accuracy

The UK government rolled out GOV.UK Chat, an Anthropic Claude powered chatbot, into the GOV.UK App on May 15, PublicTechnology reported. The app has 563,000 registered users. The system draws on 80,000 pages of official guidance from the 700,000-page GOV.UK website and claims 90% accuracy on responses to citizen queries, having switched from an earlier ChatGPT based stack.

Read at PublicTechnology ↗

Ontario auditors find 60% of approved AI medical scribes mixed up prescribed drugs in patient notes

The Ontario Office of the Auditor General audited 20 vendors approved under the province's AI Scribe program and found that 60% of the systems mixed up prescribed drugs in patient notes, The Register reported. Nine of the 20 systems fabricated treatment suggestions that were not in the underlying consultation. The AI Scribe program is an Ontario Ministry of Health initiative for clinical documentation.

Read at The Register ↗