AI Policy · Daily

Trump signaled openness to a federal equity stake in leading AI firms ahead of a planned meeting with 12 to 15 industry CEOs. Illinois lawmakers sent SB 315 to Gov. JB Pritzker, a measure requiring annual independent audits and risk disclosures from frontier AI labs starting in 2027. Anthropic pressed Congress for a rigorous federal law on catastrophic AI risks and mandatory safety tests before any preemption of state rules, while CEO Dario Amodei separately called for third-party audits of frontier models above a compute threshold and government power to block unsafe deployments. OpenAI published a threat report on two China-linked influence operations that used ChatGPT to push narratives about U.S. data center buildouts and tariffs before the company banned the accounts.

I.Top Stories

Trump publicly signals openness to federal equity stake in AI companies ahead of CEO meeting

President Trump told reporters in the Oval Office on Wednesday that he expects 12 to 15 top AI executives to agree to "giving back" to the public when he meets them shortly, an apparent reference to a federal equity stake in the firms, per Reuters. Trump told reporters the arrangement would make the public "very rich." A White House official declined to identify the executives invited to the meeting, and Anthropic, OpenAI, Google and Meta did not immediately respond to requests for comment.

Read at Investing.com ↗ Read at NYT ↗

Illinois SB 315 clears state legislature with mandatory third-party audits for frontier AI models

Illinois Senate Bill 315, which passed the state Senate 52-5 and the state House unanimously May 27, would mandate annual independent third-party audits of safety practices at frontier AI labs operating in the state, GovTech reported. Once signed by Gov. JB Pritzker, who indicated in a May 27 X post he will sign, the law would also require pre-deployment transparency reports, published catastrophic risk frameworks, critical safety incident reporting, and whistleblower protections, and would take effect Jan. 1, 2027. Sponsors and outside analysts have pitched the bill as a potential national baseline for U.S. frontier model oversight in the absence of federal AI law.

Read at GovTech ↗

Anthropic urges Congress not to override state AI laws without rigorous federal standards

Anthropic on Wednesday called on Congress not to block state AI regulations unless lawmakers pass a "rigorous" federal law addressing "catastrophic AI risks," per Reuters. The company's policy proposal also asks Congress to require AI firms to put their most powerful models through independent safety tests, and urges Congress and the states to modernize unemployment benefit technology to prepare for AI driven labor market shocks. The Hill separately reports that OpenAI and Anthropic are now working with state legislators on AI bills rather than waiting for federal action, with most labs publicly supporting a national framework that would eliminate the state patchwork.

Read at Investing.com ↗ Read at The Hill ↗

OpenAI report exposes PRC-linked influence operations targeting U.S. AI debate

OpenAI on Wednesday published a threat intelligence report describing two clusters of ChatGPT accounts, likely originating from China, that the company banned for running covert influence operations on U.S. AI policy debates. The first cluster, dubbed "Data Center Bandwagon," generated social media comments and images claiming AI data center buildouts were raising electricity prices for American families. The second, "Tech and Tariffs," criticized U.S. tariffs as attempts to dominate technological competition and amplified false claims that ChatGPT user data had been compromised, the company said. OpenAI said it found no evidence of meaningful breakout beyond the operators' own activity.

Read at OpenAI ↗

Anthropic's Amodei calls for mandatory third-party audits and government veto over dangerous AI models

Anthropic CEO Dario Amodei published an essay calling for mandatory third-party audits of frontier AI models above a compute threshold, SiliconANGLE reported. Under the proposal, governments could block deployment if independent auditors find unacceptable risk. The essay identifies four risk categories for evaluation: cybersecurity vulnerabilities, biological weapons capabilities, accelerated automated research in dangerous domains, and the potential for models to grow beyond human control. Amodei compared his framework to existing safety regimes for aircraft, automobiles and drugs, writing that AI policy must address "powerful technologies essential to the modern economy, but capable of killing large numbers of people if designed or operated poorly." The proposal goes further than Trump's June 2 voluntary AI cybersecurity executive order, which encouraged AI firms to share models with government auditors a month before public release but stopped short of mandatory pre-deployment testing.

Read at SiliconANGLE ↗ Read at Dario Amodei ↗

II.China Watch

MIIT releases 2026-2028 plan to align China's communications backbone with AI workloads

China's Ministry of Industry and Information Technology issued a three-year "AI + Information Communication" implementation plan calling for accelerated R&D on high-end optoelectronic chips, all-optical switching devices, and faster backbone transmission to feed AI compute, per Pandaily. The document targets 400 Gbps and 800 Gbps backbone speeds, photoelectric interconnects for AI supercomputing clusters, and lossless wide-area networks linking China's eastern, central and western national hub nodes. MIIT framed the plan as removing bottlenecks from chip-level components up to nationwide network architecture.

Read at Pandaily ↗

ByteDance spins off its AI drug discovery unit for independent financing

ByteDance is carving its AI drug discovery business into a standalone company and opening outside fundraising while retaining majority control, Pandaily reported. The roughly 50-person team led by Liu Kai moves to the new entity but keeps compute access via ByteDance's Volcano Engine cloud. The unit's recent output includes the Protenix protein-ligand structure prediction models released in 2025, the PXDesign protein binder tool, and an IL-17 small-molecule program disclosed in April that blocks all three IL-17 dimer isoforms with a single compound for autoimmune indications.

Read at Pandaily ↗

China's Qianfan broadband constellation crosses 200 satellites with first direct-to-cell test

China's Qianfan low-Earth-orbit constellation passed 201 satellites on Tuesday, the South China Morning Post reported, citing state broadcaster CCTV. A LandSpace Zhuque-2E rocket launched from the Gobi Desert deployed the Qianfan DTC-01 direct-to-cell test satellite alongside a China Mobile satellite. The launch followed back-to-back missions last week from Taiyuan and Hainan that added 18 satellites each. Analysts said the quickening pace still trails Qianfan's official deployment targets, and the constellation lags SpaceX's Starlink on both cost and cadence.

Read at SCMP ↗

Chinese firms showcase AI policing tools that score suspects' health and mental state at Beijing expo

Vendors at the international police and anti-terrorism technology expo in Beijing demonstrated AI enabled equipment that assesses suspects' physical condition, mental state and assigned risk levels, the South China Morning Post reported. Tianjin-based video surveillance supplier Tiandy showed a camera designed to read vital signs, including heart rate, blood pressure, blood oxygen level and blood flow rate, on up to six people at once. Exhibitors said the systems are intended to cut manpower needs and improve efficiency amid a shortage of frontline officers.

Read at SCMP ↗

III.Federal Policy Tracker

ACLU sues Florida sheriff and police over wrongful arrest tied to 93%-match AI facial recognition

Robert Dillon, a Florida man, on Tuesday filed an ACLU-backed lawsuit in U.S. district court in Fort Myers against the Jacksonville Beach Police Department, the Jacksonville Sheriff's Office, and Pinellas County Sheriff Bob Gualtieri, whose office operates the FACES facial recognition system, the Guardian reported. After a Jacksonville Sheriff's Office employee ran grainy McDonald's surveillance photos through the Pinellas-operated system, it returned a 93% match identifying Dillon as the man who attempted to lure a child, despite Dillon living more than 300 miles away in Fort Myers. The charges were dismissed last year over a November 2023 incident for which Dillon was arrested in August 2024, and the suit alleges his case is at least the 15th nationally involving someone charged or arrested after a false AI face match. In the lawsuit, the ACLU said "the investigation resulted in the wrongful arrest and prosecution of an innocent man."

Read at Guardian ↗

New York AI disclosure law for advertising takes effect

A New York state law requiring anyone who creates advertising in the film and television industry to disclose if their commercials include AI generated synthetic performers took effect Tuesday, per WRGB. Gov. Kathy Hochul's office said advertisements without a proper label can undermine consumers' ability to distinguish fact from fiction. Hochul, announcing the law is in effect, said requiring disclosure when an ad uses synthetic performers "protects consumers, respects our creative workforce and keeps New York at the forefront of responsible innovation."

Read at 13WHAM ↗ Read at NY Governor's Office ↗

Maryland names Michael Boyce as senior adviser for responsible AI

Gov. Wes Moore on Wednesday named Michael Boyce as Maryland's senior adviser for responsible AI, GovTech reported. Boyce previously led the AI portfolio program at U.S. Digital Response, served as director of the AI Corps at the Department of Homeland Security, and wrote generative AI policy at the White House Office of Management and Budget under the 2023 AI executive order. His hiring follows Maryland's launch of an AI Innovation Lab to support agency adoption, and he succeeds Nishant Shah, whose final day was Jan. 16.

Read at GovTech ↗

IV.Capability & Research Watch

Microsoft ships 200-plus security patches with AI driven bug discovery cited as cause

Microsoft on Tuesday released more than 200 fixes for security vulnerabilities, 38 of them critical, up from the prior record of 175 fixes set in October, SiliconANGLE reported. One zero-day flaw in HTTP.sys, tracked as CVE-2026-49160, was credited to OpenAI's Codex, one of the early publicly attributed cases of an AI system reporting a vulnerability in a major Patch Tuesday cycle. The volume follows Microsoft's AI driven discovery push, including MDASH, an agentic scanning system with more than 100 AI agents that surfaced 16 previously unknown flaws patched in May. Dustin Childs, head of threat awareness at Trend Micro's Zero Day Initiative, told SiliconANGLE that AI is "supercharging flaw discovery at an uncontrollable scale."

Read at SiliconANGLE ↗ Read at Microsoft ↗

Anthropic reverses Claude Fable 5 covert sabotage feature for AI researchers after backlash

Anthropic on Wednesday reversed a Claude Fable 5 policy that would have covertly degraded the model's performance for users attempting to train competing AI systems, telling Wired the company "made the wrong trade-off and we apologize for not getting the balance right." Safeguards for frontier AI development will now be visible: Claude Fable 5 will either refuse the request or reroute the user to a less capable model, rather than silently degrade output. The original detail had been buried in Anthropic's system card and surfaced after research community pushback, including from Dean Ball, a senior fellow at the Foundation for American Innovation and a former White House AI adviser, who called the policy "shockingly hostile." Anthropic separately said it would grant safeguard-free access to a scientific community for legitimate research.

Read at Wired ↗

V.Industry & Market Watch

OpenAI, Nvidia and Anthropic boosted H-1B applications despite Trump's $100,000 fee; federal judge struck the charge down

OpenAI more than tripled its certified H-1B applications and Nvidia's grew 19% in the first quarter of 2026 compared with 2025, even at $100,000 per new petition, while Anthropic's filings grew from roughly 10 to nearly 60, Fortune reported. Over the same period, Amazon, Google and Microsoft posted steep declines, with smaller dips at Meta and Apple. On Monday, U.S. District Judge Leo Sorokin in Boston ruled the fee an unauthorized tax, writing that the policy "imposes a tax on H-1B petitions without the requisite delegation by Congress." A separate D.C. federal court upheld the fee in a U.S. Chamber of Commerce challenge that is now on appeal. The question is likely headed to a higher court even as the fee is set to expire on its own in September 2026.

Read at Fortune ↗

Visa and OpenAI announce AI agent payment integration for ChatGPT

Visa and OpenAI on Wednesday announced an integration that lets AI agents inside ChatGPT initiate Visa transactions on users' behalf, SiliconANGLE reported. Payments run within user-defined permissions including spending caps, merchant categories and required approvals. Visa will supply tokenization, real-time authorization, fraud monitoring, and chargeback handling. The companies said they will also explore enterprise uses, including allowing OpenAI's Codex coding agent to buy inference, application programming interfaces or other developer services on its own within limits set by users. Visa's global head of growth Rubail Birwadker told Axios that more than one in five transactions are now being influenced by what consumers learn from large language models.

Read at SiliconANGLE ↗ Read at Visa ↗

VI.Global & Geopolitics

Canada introduces Digital Safety Act banning under-16s from social media and regulating AI chatbots

Canadian Culture Minister Marc Miller on Wednesday introduced the Digital Safety Act, per SCMP. The bill would ban social media accounts for children under 16 and require AI chatbot services to mitigate harmful content and label synthetically generated content. Platforms could earn an exemption pathway by demonstrating sufficient safeguards for children. A new Digital Safety Commission would enforce the rules with fines up to 3% of global revenue or C$10 million (U.S. $7.2 million) for noncompliance. Miller said in the announcement that "the safety of children cannot be an afterthought."

Read at SCMP ↗