AI Policy · Daily

Anthropic turned down Beijing's bid for its newest Mythos model as Trump heads to the Xi summit. House Oversight Republicans launched a probe of Sam Altman's outside investments, joined by six state attorneys general pressing the SEC ahead of OpenAI's IPO. Commerce and the intelligence community are clashing inside the Trump administration over who evaluates frontier AI models before release. FOIA records show DOGE pitched HUD an AI tool called SweetREX built to recommend deleting regulations by default.

I.AI Policy Today

Anthropic refuses Chinese request for access to its newest AI model on eve of Trump-Xi summit

Anthropic confirmed it refused a Chinese request for access to its newest model Mythos, the New York Times reported. The Times framed the denial as evidence that the latest Anthropic and OpenAI models are extending the U.S. lead over China at the frontier model layer. The report lands the morning President Trump is set to depart for the May 14-15 Beijing summit with Xi Jinping, where AI joins the Strait of Hormuz and Taiwan as a top three agenda item, as reported by NPR in AIPD's May 11th edition.

Read at NYT ↗

House Oversight opens probe of Sam Altman's investments; six GOP state AGs urge SEC review before OpenAI IPO

The Republican-led House Oversight Committee opened an investigation into OpenAI Chief Executive Sam Altman's business dealings, the Wall Street Journal reported. The committee sent the company a letter requesting documents on governance practices and potential conflicts tied to Altman backed companies. Six Republican state attorneys general from Florida, Montana, Nebraska, Iowa, West Virginia and Louisiana urged the Securities and Exchange Commission to review OpenAI's governance before any IPO, per Investing.com summarizing the WSJ report. The scrutiny follows prior WSJ reporting on Altman's efforts to have OpenAI support companies he personally invested in, including nuclear fusion startup Helion and aerospace firm Stoke Space. OpenAI board chairman Bret Taylor defended Altman during a Monday court hearing in the Musk v. Altman trial, saying the chief executive had been forthright about outside investments.

Read at WSJ ↗ • Read at Investing.com ↗

Trump administration splits over expanded intelligence community role in AI model evaluations

An internal Trump administration fight pits the Commerce Department against intelligence agencies over an expanded IC role in evaluating frontier AI models, the Washington Post reported. Reporters Cat Zakrzewski, Ellen Nakashima and Nitasha Tiku cite named sources and reference internal documents, noting that a related website was removed amid White House sensitivities. The dispute follows Commerce's CAISI signing binding pre-release testing pacts with Google DeepMind, Microsoft and xAI, as reported by NIST in AIPD's May 6th edition.

Read at Washington Post ↗

FOIA documents reveal DOGE pitched a HUD AI tool ("SweetREX") to recommend rule deletions by default

Newly disclosed FOIA documents obtained by Democracy Forward show DOGE delivered a PowerPoint last summer pitching an AI tool called SweetREX at the Department of Housing and Urban Development, FedScoop reported. The tool, named for DOGE associate Christopher Sweet, would analyze every HUD regulation and recommend keeping, deleting or partially deleting each one, with deletion presented as "the default case." The HUD regulations covered include the prohibition of sex-based discrimination in mortgage assistance and foreclosure legal aid. The disclosure was first reported by The Lever and shared with FedScoop. The pitch told HUD employees the tool would "automate all the most time-consuming steps in deregulation" while leaving staff in control throughout the process.

Read at FedScoop ↗

FTC Chair Ferguson warns 15 tech companies to comply with Take It Down Act by May 19

Federal Trade Commission Chairman Andrew Ferguson sent letters to 15 technology companies, including Amazon, Alphabet, Apple, Automattic, Bumble, Discord, Match Group, Meta, Microsoft, Pinterest, Reddit, SmugMug, Snapchat, TikTok and X, reminding them of obligations under the Take It Down Act with a May 19 compliance deadline, per an FTC press release. The law requires covered platforms to take down nonconsensual images, and all identical copies, within 48 hours of receiving a valid request. The Take It Down Act, signed last year by President Trump with the support of First Lady Melania Trump, targets nonconsensual intimate imagery including AI generated deepfakes. Ferguson said the agency stands "ready to monitor compliance, investigate violations, and enforce the Take It Down Act."

Read at FTC ↗

Wrongful death suits against OpenAI test consumer product safety theory to bypass Section 230

Plaintiff lawyers in pending wrongful death suits against OpenAI are advancing a consumer product safety legal theory that would expose chatbot makers to product defect liability standards, the New York Times reported. The theory seeks to apply existing consumer product safety statutes to chatbot companies for harms tied to user suicides and violent acts. A favorable ruling would reframe chatbot liability around product defect doctrine rather than speech or Section 230. The argument extends seven complaints over the Tumbler Ridge mass shooting filed April 29 in the U.S. District Court for the Northern District of California. The complaints allege GPT-4o was a defective product that failed to challenge the shooter or refer her flagged account to law enforcement, per OPB.

Read at NYT ↗ • Read at OPB ↗

II.China Watch

SMIC clears Shanghai Stock Exchange review for $5.9 billion buyout of fab subsidiary

Semiconductor Manufacturing International Corp. said Monday that the Shanghai Stock Exchange's M&A Review Committee approved its plan to issue shares and acquire the remaining 49% stake in subsidiary SMIC North, TechNode reported. The transaction is valued at 40.6 billion yuan ($5.9 billion) and pays out five minority shareholders including the state-backed China Integrated Circuit Industry Investment Fund, known as the Big Fund. SMIC North runs 12-inch wafer lines at 28nm and above with monthly capacity of 75,000 wafers, making the deal the largest M&A in China's domestic foundry industry to date. The approval advances Beijing's push to consolidate domestic chipmaking under fewer national champions as U.S. export controls tighten access to leading-edge equipment, locking in a Big Fund exit at the same time.

Read at TechNode ↗

Kuaishou seeks $20 billion valuation for Kling AI video spin-off; Tencent in talks

Chinese short-video company Kuaishou plans to carve its Kling video generation business out for a separate listing at a $20 billion valuation, with about $2 billion in pre-IPO funding under discussion, per QbitAI, citing The Information. Tencent has appeared on the list of potential investors, the report added, citing LatePost. Kling's annualized run-rate revenue exceeded $300 million in January and reached roughly $500 million by the end of April. At a $20 billion price, Kling would equal about 70% of Kuaishou's Hong Kong market capitalization.

Read at QbitAI ↗

360 Digital Security reports 23 AI agent vulnerabilities through China's national disclosure platforms

360 Digital Security Group published an "OpenClaw Ecosystem Security Risk Analysis" report Tuesday on 23 vulnerabilities uncovered across OpenClaw's core and 10 derivative AI agent products, QbitAI reported. The flaws span remote code execution, authentication bypass, privilege escalation and information disclosure, and all findings were submitted to the China National Vulnerability Database (CNNVD) and the China National Vulnerability Sharing Platform (CNVD). The firm attributed the discoveries to its in-house vulnerability mining agent running automated audits across the agent supply chain. QbitAI noted OpenClaw's GitHub has cumulatively logged more than 535 security advisories, with new advisories averaging four or more per day since the first quarter of 2026.

Read at QbitAI ↗

State media says AI driven "dark factory" more than doubled J-20 stealth fighter production efficiency

China has more than doubled production efficiency for J-20 stealth fighter components in a "dark factory" running autonomous vehicles and AI driven machinery nearly 24 hours a day, according to the official Science and Technology Daily (state media) as cited by the South China Morning Post. The plant can now assemble the "skeleton" of an aircraft in near-darkness, work that previously required round-the-clock human monitoring. The state outlet framed the throughput gain as proof that AI is lifting China's industrial base. The plant is operated by Chengdu Aircraft Corporation, a subsidiary of state-owned Aviation Industry Corporation of China, with digital manufacturing center chief Song Ge telling the outlet that the J-20 contains thousands of unique components.

Read at SCMP ↗

III.Industry & Market Watch

California AG Bonta announces $12.75 million CCPA settlement with GM over sale of OnStar driver data

California Attorney General Rob Bonta announced a $12.75 million proposed settlement with General Motors for alleged California Consumer Privacy Act violations tied to the sale of California drivers' driving and location data, BleepingComputer reported. The allegations cover 2020 to 2024 sales through OnStar and Smart Driver to Verisk Analytics and LexisNexis Risk Solutions, where the data fed insurance-related driver-scoring products and generated about $20 million in revenue nationwide. The Federal Trade Commission previously banned GM from selling drivers' data for five years.

Read at BleepingComputer ↗

Santa Clara County sues Meta over billions of scam advertisements on Facebook and Instagram

Santa Clara County, California filed suit Monday in Santa Clara County Superior Court against Meta Platforms, alleging the company knowingly facilitates and profits from billions of scam advertisements across Facebook and Instagram in violation of California's false advertising and unfair business practices laws, Bloomberg reported. The complaint, citing earlier Reuters reporting, alleges Meta earned as much as $7 billion in annual revenue from high risk scam ads, used algorithmic targeting to push more scam ads at users who had clicked similar offerings, and allowed flagged advertisers to keep running ads at higher penalty-bid prices, per Reuters. The county further alleges Meta's generative AI systems often assist unethical marketers in creating scam content. County Counsel Tony LoPresti is working with three outside law firms on contingency: Bernstein Litowitz Berger & Grossmann; Renne Public Law Group; and Bishop Partnoy.

Read at Bloomberg ↗ • Read at Reuters ↗

SoftBank's Son in talks with Macron on multi-billion-dollar AI data center project in France

SoftBank Group founder Masayoshi Son has held talks with President Emmanuel Macron about unveiling a French AI data center project in the coming weeks, Bloomberg reported, citing people familiar with the matter. Son has floated as much as $100 billion in potential French investment, though the actual figure could come in materially lower if he prioritizes other projects, the people said. Macron and Son may announce plans during the Choose France Summit, an annual investment gathering. The French talks follow SoftBank's March announcement of a potential $500 billion Ohio data center buildout and SoftBank's commitment of more than $60 billion to OpenAI for a roughly 13% stake.

Read at Bloomberg ↗ • Read at Yahoo Finance ↗

IV.Global & Geopolitics

South Korean presidential policy chief Kim Yong-beom proposes AI "citizen dividend" funded by chipmaker excess profits

Kim Yong-beom, chief of the South Korean Presidential Policy Office, proposed on Facebook a "citizen dividend" funded by taxes on excess profits from the AI and semiconductor boom, saying gains from the AI infrastructure era are not solely the result of individual companies, Bloomberg reported. Samsung Electronics and SK Hynix shares dipped on the comments and Korean stocks swung intraday. Kim raised potential uses including startup funding for young entrepreneurs, basic income for rural and fishing communities, stronger pensions for older people and AI era education accounts, per Korea JoongAng Daily. The Presidential Office said the proposal is "a personal opinion unrelated to internal discussions or reviews" inside the Blue House.

Read at Bloomberg ↗ • Read at Korea JoongAng Daily ↗

Paris prosecutors seek summons of Elon Musk, xAI, X Corp. and former CEO Yaccarino over Grok linked deepfakes and CSAM

The Paris public prosecutor's office has requested proceedings be initiated against xAI Holdings Corp., X Corp., xAI, Elon Musk and former X chief executive Linda Yaccarino "by summoning them … or, in the event of non-compliance, by issuing a warrant equivalent to an indictment," Biometric Update reported. The probe covers sexualized deepfakes, alleged complicity in possession and distribution of child sexual abuse material linked to Grok, and possible unlawful data processing. Le Monde flagged a financial-motive angle and reported that Musk publicly mocked French authorities. The action escalates the French criminal investigation into Musk and X first reported by Le Monde in AIPD's May 8th edition.

Read at Biometric Update ↗

OpenAI offers EU vetted access to new cyber model; Brussels says Anthropic still negotiating on Mythos

OpenAI announced Monday it was granting preview access to its latest cyber model to vetted EU cybersecurity teams, CNBC reported. European Commission spokesperson Thomas Regnier said the Commission has received OpenAI's offer while four or five meetings with Anthropic have not yet produced any path to Mythos access for the bloc, per Reuters. Former British finance minister George Osborne, who heads OpenAI's "OpenAI for Countries" initiative, sent a letter to the Commission and member states laying out the company's EU Cyber Action Plan. The letter came a month after the European Commission said ChatGPT should be regulated as a large online search engine under the Digital Services Act.

Read at CNBC ↗ • Read at Reuters ↗

Australia's AUSTRAC warns money launderers are using AI to scale up fraud and forge documents

Australia's financial crimes watchdog AUSTRAC warned in updated annual risk guidance that criminals are increasingly using AI to scale up activities, automate processes and create fake documents, Bloomberg reported. AUSTRAC chief executive Brendan Thomas said criminals are increasingly fabricating identities, forging documents and disguising the proceeds of scams using AI. AUSTRAC released the AI guidance alongside revised national risk assessments for money laundering, terrorism financing and proliferation financing tied to Iran and North Korea. From July 1, real estate agents, dealers in precious metals and stones, and trust and company service providers come under new AML/CTF rules under Australia's Tranche 2 reforms.

Read at Bloomberg ↗ • Read at The Star ↗