AI Policy · Daily

Lieu and Obernolte unveiled a bipartisan AI omnibus Monday that codifies more than 20 House AI Task Force recommendations, funds the AI Safety Institute consortium at $10 million and omits state preemption language. Microsoft and OpenAI rewrote their partnership terms, dropping the AGI termination trigger, converting the IP license to nonexclusive through 2032 and freeing OpenAI to ship across rival clouds including Amazon Bedrock. Jurors were seated in Oakland for the Musk v. Altman trial opening today, with the plaintiff seeking up to $134 billion and the removal of Sam Altman and Greg Brockman over OpenAI's conversion to a public benefit corporation.

I.AI Policy Today

Reps. Lieu and Obernolte introduce bipartisan American Leadership in AI Act with explicit no-preemption stance

Reps. Ted Lieu (D-Calif.) and Jay Obernolte (R-Calif.), co-chairs of the House Bipartisan AI Task Force, introduced the American Leadership in AI Act on Monday, codifying more than 20 of the task force's recommendations from its December 2024 final report. The bill funds the Consortium for AI Safety Institute at $10 million, establishes the National AI Research Resource, creates an AI Workforce Research Hub, expands whistleblower protections and authorizes civil rights actions against deepfakes and fraud. The text declines to preempt state AI developer regulations, contrasting with Energy and Commerce Chair Brett Guthrie's SECURE Data Act and the Trump December executive order's preemption mandate. The bill is the first bipartisan AI omnibus introduced in the 119th Congress.

Read at Punchbowl News ↗

Microsoft and OpenAI scrap AGI exclusivity clause and convert license to nonexclusive through 2032

Microsoft and OpenAI announced Monday that they had revised the contract governing their partnership. The new terms drop the clause that allowed OpenAI to terminate Microsoft's access upon declaring artificial general intelligence and convert Microsoft's intellectual property license to a nonexclusive arrangement running through 2032. OpenAI may now serve its products across any cloud provider, with Amazon chief executive Andy Jassy saying ChatGPT models will be available on Amazon Bedrock within a few weeks alongside a new Stateful Runtime Environment for AI agents. Microsoft's revenue share from OpenAI continues at the same percentage but is now capped through 2030, while Microsoft will no longer pay OpenAI a share of its AI revenue. Microsoft retains right of first ship on Azure unless it cannot and chooses not to support a given capability.

Read at The Verge ↗ Read at SiliconANGLE ↗

Opening arguments begin in Musk v. Altman with nine jurors seated and $134 billion in damages on the table

Judge Yvonne Gonzalez Rogers seated nine jurors Monday in U.S. District Court in Oakland, Calif., opening three weeks of testimony in Musk v. Altman starting today. Musk is seeking up to $134 billion in damages from OpenAI and Microsoft and asking the court to remove Sam Altman and Greg Brockman and restore OpenAI as a nonprofit. Witnesses are expected to include Musk, Altman, Brockman, former chief scientist Ilya Sutskever, former chief technology officer Mira Murati and Microsoft chief executive Satya Nadella. The jury will deliver an advisory verdict to guide the judge on Musk's claims, which target OpenAI's October 2025 conversion from nonprofit to public benefit corporation.

Read at MIT Technology Review ↗ Read at CNBC ↗

OpenAI clears FedRAMP Moderate, opening federal procurement on-ramp closed to Anthropic

OpenAI announced Monday that ChatGPT Enterprise and the OpenAI API received FedRAMP Moderate authorization, allowing federal agencies to procure both under standard government cloud security terms. The authorization clears the procurement gate for most civilian agencies and arrives as Anthropic remains locked out of Pentagon contracts after the Department of Defense designated the company a supply chain risk over its refusal to license Claude for "all lawful purposes." OpenAI subsequently revised its own Pentagon language to bar deliberate tracking, surveillance or monitoring of U.S. persons.

Read at OpenAI ↗

More than 600 Google and DeepMind staff urge Pichai to refuse classified Pentagon AI work, citing Anthropic precedent

More than 600 Google and Google DeepMind employees signed an open letter on Monday urging chief executive Sundar Pichai to refuse any classified workloads in ongoing Department of Defense negotiations over Gemini, The Hill reported. The letter follows the Pentagon's designation of Anthropic as a supply chain risk earlier this year after Anthropic refused the "all lawful purposes" clause. Signatories include more than 20 directors, senior directors and vice presidents across Google, alongside a number of senior employees at Google DeepMind, with the letter saying rejection of classified workloads is the only way to prevent surveillance and weapons applications.

Read at The Hill ↗ Read at Bloomberg ↗

DOJ unseals charges against Silk Typhoon hacker Xu Zewei after extradition from Italy in named MSS-directed campaign

The Department of Justice on Monday announced the extradition of Chinese national Xu Zewei, who arrived from Italy over the weekend and made his initial appearance in federal court in Houston, BleepingComputer reported. The indictment alleges Xu worked for Shanghai Powerock Network at the direction of the Ministry of State Security's Shanghai State Security Bureau and ties him to the Silk Typhoon (Hafnium) campaign that exploited Microsoft Exchange Server zero-days from late 2020 onward, compromising more than 12,700 U.S. organizations. The campaign also targeted COVID-19 research organizations, with prosecutors alleging the operators sought data on vaccines, treatments and testing. Xu pleaded not guilty on a nine-count indictment that includes wire fraud, computer intrusion and aggravated identity theft.

Read at BleepingComputer ↗

Senate Armed Services holds closed SOCOM and CYBERCOM posture hearing today on FY27 defense request

The Senate Armed Services Committee holds a closed hearing this morning to review the posture of U.S. Special Operations Command and U.S. Cyber Command in connection with the Defense Authorization Request for fiscal year 2027 and the Future Years Defense Program. The closed session is followed by an open session at 11 a.m. in SD-G50. The hearing falls within the same week the Pentagon's $1.5 trillion FY27 budget request lands on Capitol Hill. The request earmarks roughly $54 billion for the Defense Autonomous Warfare Group, the Pentagon's reorganized focal organization for autonomous and drone capabilities.

Read at Senate Armed Services ↗

II.China Watch

NDRC blocks Meta's $2 billion Manus acquisition in first AI sector veto under 2020 framework

China's National Development and Reform Commission on April 27 prohibited Meta's $2 billion acquisition of Manus parent Butterfly Effect and ordered the parties to unwind the closed deal, TechNode reported. The Foreign Investment Security Review Office issued the harshest available ruling under the 2020 Foreign Investment Security Review Measures, marking the first publicly disclosed AI deal block under that framework. Manus chief executive Xiao Hong and chief scientist Ji Yichao were reportedly barred from leaving China during the review, said the report. Beijing tied the decision to growing scrutiny on AI exports. The ruling gives Treasury and Commerce officials direct precedent as they weigh reciprocal CFIUS reforms and outbound investment rules now under review.

Read at TechNode ↗

DeepSeek V4 release shows core R&D talent migrating to ByteDance and Tencent

DeepSeek's 58-page V4 technical report listed 10 contributors among nearly 300 authors as having departed the company, TechNode reported. At least five core R&D members have left since the second half of 2025, covering base models, reasoning, OCR and multimodal research, said the report. Identified departures include Guo Daya, a core author of DeepSeek-R1, and Wang Bingxuan, a core author of DeepSeek's original large language model, with both reportedly recruited to ByteDance and Tencent. DeepSeek opened external financing in mid-April to retain remaining staff and fund larger scale training. The redistribution of frontier model talent into ByteDance and Tencent gives Commerce and BIS officials specific firms to factor into pending compute and cloud access control reviews.

Read at TechNode ↗

Xiaomi says 3nm Xuanjie O1 chip passes 1 million shipments

Xiaomi chief executive Lei Jun said Monday that the company's in-house 3nm flagship Xuanjie O1 has surpassed 1 million shipments since its May 2025 launch, TechNode reported. Lei told an investor day event that Xiaomi has become the fourth company globally capable of independently developing high-end smartphone system-on-chips. The Xuanjie series ships in the Xiaomi 15S Pro, Pad 7 Ultra and 7S Pro, and will extend to Xiaomi electric vehicles and other smart devices with annual upgrades, said the report. Existing BIS controls cover U.S. origin tools and software but not chip design itself, leaving Chinese fabless designers free to use 3nm fabrication capacity abroad.

Read at TechNode ↗

Yuanjie Semiconductor's Q1 profit surges 1,153% on AI optical chip demand

Chinese optical chipmaker Yuanjie Semiconductor Technology reported Q1 net profit up 1,153% year on year to 179 million yuan, about $26 million, with revenue up 321% to 355 million yuan, the South China Morning Post reported. The Shanghai-listed company makes laser chips for high-speed data transmission used in AI data centers, 5G infrastructure and fiber access networks. Yuanjie shares are up about elevenfold over the past year, closing at 1,418 yuan on Monday and topping Kweichow Moutai as mainland China's costliest stock. Yuanjie's revenue trajectory is a concrete data point for Commerce officials reviewing optical interconnect and photonic chip controls, relevant to pending decisions on whether and how to expand the export control perimeter.

Read at SCMP ↗

III.Federal Policy Tracker

DOJ extends ADA Title II web accessibility deadline by one year, citing generative AI's failure to automate remediation at scale

The Department of Justice published an interim final rule extending the Americans with Disabilities Act Title II compliance deadline for public entities with populations over 50,000 from April 24, 2026 to April 26, 2027, with smaller entities pushed from April 26, 2027 to April 26, 2028, THE Journal reported. The rule text states that "advanced technology, such as generative AI, does not yet reliably automate the remediation of inaccessible content at scale," directly attributing part of the extension to AI capability gaps. Public comments on the interim final rule are split, with disability advocates urging no further delay and arguing institutions have had sufficient time. The rule is one of the few instances of an agency naming generative AI capability shortfalls in rulemaking text.

Read at THE Journal ↗

Alabama names Aaron Wright as state's first chief AI officer

Alabama promoted Aaron Wright, the state's director of application development, to chief artificial intelligence officer, with his first day Monday, StateScoop reported. The appointment follows a 2024 executive order from Gov. Kay Ivey establishing a generative AI task force charged with surveying agency use of AI and recommending policy guidance. The task force's first report found nearly a quarter of Alabama agencies were already using generative AI for text generation, language translation, software code development and complex problem-solving. Wright led the task force's working group on data management and ownership during a four-year tenure at the Alabama Office of Information Technology.

Read at StateScoop ↗

Deloitte-NASCIO survey finds state CISO confidence has fallen by half since 2022 amid AI driven threats

The biennial Deloitte-NASCIO survey of all 50 states plus the District of Columbia and the Virgin Islands found the share of state chief information security officers reporting they are extremely or very confident in their cyber defenses fell from 48% in 2022 to 22% in 2026, StateScoop reported. Sixteen percent of states reported cyber budget cuts this cycle, compared with zero in 2024. CISOs cited AI driven phishing and ransomware-as-a-service as the top emerging threats. The Cybersecurity and Infrastructure Security Agency itself operates without a Senate-confirmed director after Sean Plankey withdrew his nomination last week.

Read at StateScoop ↗

IV.Industry & Market Watch

Google signs classified Pentagon AI deal with "any lawful government purpose" language, agrees to adjust safety filters at government request

Google signed an amendment to its existing Department of Defense contract Monday allowing the Pentagon to use its AI for classified workloads under "any lawful government purpose" terms, The Information reported, hours after more than 600 Google and DeepMind employees urged chief executive Sundar Pichai to reject the agreement. The terms run more permissive than OpenAI's February Pentagon deal: where OpenAI retained "full discretion over our safety stack," Google's contract requires the company to assist in adjusting AI safety settings and filters at the government's request. The agreement includes language stating the AI "is not intended for, and should not be used for, domestic mass surveillance or autonomous weapons (including target selection) without appropriate human oversight," but adds that the deal "does not confer any right to control or veto lawful Government operational decision-making." Charlie Bullock, senior research fellow at the Institute for Law and AI, told The Information that the "is not intended for" framing is "not legally binding in any way." Google joins xAI and OpenAI in classified Pentagon AI deployments, leaving Anthropic as the only frontier lab outside DoD classified workloads after its February supply chain risk designation, which the company is currently challenging in federal court.

Read at The Information ↗

Anthropic implied valuation crosses $1 trillion in private secondary trading

Anthropic's implied pre-IPO valuation crossed $1 trillion in on-chain private market trading on the Jupiter decentralized platform, the Economic Times reported. The pre-IPO instruments are backed by special purpose vehicle exposure to Anthropic shares, giving secondary buyers a live valuation read ahead of any public listing. Reported revenue rose from roughly $9 billion at the end of 2025 to over $30 billion by March, with Alphabet preparing an additional investment of up to $40 billion and Amazon committed to $25 billion plus a long-term cloud agreement. Anthropic's secondary trading at trillion-dollar levels arrives as Sen. Josh Hawley (R-Mo.) presses Republicans on AI industry lobbying.

Read at Economic Times ↗

Ineffable Intelligence raises $1.1 billion seed at $5.1 billion valuation with UK Sovereign AI Fund participation

British startup Ineffable Intelligence raised $1.1 billion in seed funding at a $5.1 billion valuation, SiliconANGLE reported. Lightspeed Venture Partners and Sequoia Capital led the round, joined by Nvidia, Google, the UK Sovereign AI Fund, DST Global and Index Ventures. Founder David Silver, formerly of Google DeepMind and lead developer of AlphaGo, plans to build what the company calls a "superlearner" using reinforcement learning techniques without conventional pre-training. UK Sovereign AI Fund participation puts a foreign government investor on the cap table of a U.S.-led frontier AI seed round, a structure CFIUS staff are tracking as outbound and inbound AI investment frameworks remain under interagency review.

Read at SiliconANGLE ↗

OpenAI missed internal user and revenue targets, fueling internal concern over infrastructure spend

OpenAI fell short of its own user-acquisition and sales goals in recent months, raising concerns among company leaders about whether it can sustain its massive data-center spending, the Wall Street Journal reports. CFO Sarah Friar has told colleagues she is worried OpenAI may not be able to pay for future compute contracts if revenue growth lags, while the board has more closely scrutinized CEO Sam Altman's dealmaking spree, putting Friar at odds with Altman who has favored a more aggressive IPO timeline. OpenAI missed its goal of one billion weekly active ChatGPT users by year-end 2025, missed its annual ChatGPT revenue target as Google's Gemini gained share, and missed multiple monthly revenue targets earlier this year after losing ground to Anthropic in coding and enterprise. The company carries roughly $600 billion in future compute commitments against a recent $122 billion funding round it expects to burn through within three years assuming ambitious revenue targets are met. A material gap between revenue plans and compute commitments at the largest U.S. frontier lab is a data point for SEC and Treasury staff watching pre-IPO disclosures and the systemic exposure of cloud and chip suppliers.

Read at WSJ ↗

OpenAI missed internal user and sales targets, fueling internal concern over infrastructure spend

OpenAI missed its own user-acquisition and sales goals in recent months, fueling internal concern that the company may struggle to support its AI infrastructure spending, the Wall Street Journal reported, per Bloomberg. The miss arrives as OpenAI is restructuring its Microsoft partnership, preparing for a public offering, and committing tens of billions of dollars to compute capacity. Microsoft's revenue share from OpenAI continues at the same percentage but is now capped through 2030 under the revised partnership. A material gap between revenue plans and compute commitments at the largest U.S. frontier lab is a data point for SEC and Treasury staff watching pre-IPO disclosures and the systemic exposure of cloud and chip suppliers.

Read at Bloomberg ↗

Rural anti-AI organizing intensifies across Indiana and Idaho data center sites

Grassroots opposition to AI development has emerged across states from Indiana to Idaho, framing the conflict around Big Tech profits versus local costs in data centers, jobs and energy, the New York Times reported Monday. Rural communities are mobilizing against AI infrastructure siting, putting them at odds with the Trump administration's pro-AI infrastructure stance, the Financial Times separately reported. The opposition is intensifying in the same Idaho region where the Department of Energy opened its DOME microreactor test bed at Idaho National Laboratory earlier this month.

Read at NYT ↗ Read at FT ↗

V.Global & Geopolitics

European Commission unveils proposals to open Android ecosystem to rival AI services

The European Commission unveiled a slate of proposals on Monday aimed at opening Google's Android ecosystem to rival AI services, Bloomberg reported. The action targets Google's distribution control over AI assistants on Android phones, building on the Digital Markets Act framework. The Commission framed the proposals as a remedy to Google's control over which AI services reach Android users by default. The Department of Justice is running post-trial remedy proceedings against Google in the federal search antitrust case on a parallel U.S. track.

Read at Bloomberg ↗

HMRC rolls out 28,000 Microsoft Copilot licenses across the UK tax authority

HM Revenue and Customs deployed about 28,000 Microsoft Copilot licenses to staff and is preparing to switch on agentic features inside "Official Sensitive" workflows, The Register reported. Chief AI Officer James Mitton said at the Think AI for Government event that HMRC plans to become "the most AI-enabled tax authority on the planet." A prior 20,000-civil servant Government Digital Service trial across a dozen UK departments reported an average time saving of 26 minutes per user per day, with 70% reporting reduced time on information searches and routine tasks. The rollout is the largest known production deployment of Microsoft Copilot inside an allied tax administration.

Read at The Register ↗

South Africa pulls draft national AI policy after AI generated citations turned out to be fabricated

South Africa's Department of Communications and Digital Technologies withdrew its draft national AI policy after News24 reported that at least six references in the document were fabricated, The Register reported. Communications Minister Solly Malatsi said the department rechecked the draft and found "various fictitious sources" in the reference list, prompting withdrawal and consequence management for those involved. The draft had already cleared Cabinet and was out for public comment when the fabricated citations were identified. The case parallels a 2025 Australian incident in which Deloitte had to clean up a government report after AI generated citations slipped into a federal deliverable.

Read at The Register ↗