AI Policy · Daily

  • Kratsios elevates Chinese AI distillation to "industrial-scale theft" in a White House memo, hardening the administration's rhetoric the same week House Foreign Affairs marks up bipartisan export controls on the same practice.
  • Johnson's Section 702 text strips the warrant requirement privacy hawks demanded and dares Democrats to extend surveillance authority with Kash Patel running the FBI, six days before the April 30 expiration.
  • Southcom becomes the first combatant command with a standing autonomous warfare organization, operationalizing the $54 billion FY27 Defense Autonomous Warfare Group line before Congress has appropriated it.
  • DeepSeek previewed its V4 model one year after R1 rattled Silicon Valley, billing it as the most powerful open source AI platform.

I.AI Policy Today

OSTP Director Kratsios accuses Chinese entities of "industrial-scale" theft of U.S. AI technology

White House Office of Science and Technology Policy Director Michael Kratsios issued a memo Thursday accusing Chinese entities of running "industrial-scale" campaigns to distill frontier U.S. AI models, the Financial Times reported. The Trump administration pledged to crack down on Chinese firms exploiting U.S. AI, NPR reported. The memo lands the same week as the House Foreign Affairs Committee's bipartisan export control markups targeting Chinese AI model distillation, as reported by Bloomberg in AIPD's April 23rd edition. The memo does not specify which agencies will lead the enforcement effort.

Read at FT ↗ • Read at NPR ↗

Speaker Johnson unveils new Section 702 text without warrant requirement

Speaker Mike Johnson released new Section 702 reauthorization text Thursday extending the foreign intelligence surveillance authority for three years, Roll Call reported. The proposal omits the warrant requirement for U.S.-person queries that the bipartisan privacy bloc had demanded. Minority Leader Hakeem Jeffries said Democrats may withhold support while Kash Patel leads the FBI. The statutory authority expires April 30, 2026.

Read at Roll Call ↗

Southcom commander directs formal stand-up of Autonomous Warfare Command

Marine Corps Gen. Francis L. Donovan directed the formal establishment of a Southcom Autonomous Warfare Command Thursday, DOD News reported. The new organization will "employ autonomous, semiautonomous and unmanned platforms and systems to counter threats and challenges across the region." Southcom is the first U.S. combatant command to stand up a standing autonomous warfare organization. The move follows the Pentagon's FY27 budget request of roughly $54 billion for the Defense Autonomous Warfare Group, as reported by The Guardian in AIPD's April 23rd edition.

Read at DOD News ↗

CISA orders federal patching of "BlueHammer" Microsoft Defender zero-day

The Cybersecurity and Infrastructure Security Agency added a Microsoft Defender privilege escalation flaw dubbed "BlueHammer" to its Known Exploited Vulnerabilities catalog Monday, BleepingComputer reported Thursday. The binding directive compels federal civilian executive branch agencies to patch by May 7. The vulnerability was exploited in zero-day attacks before Microsoft issued the patch on April 14.

Read at BleepingComputer ↗

UK opens government-level talks with Anthropic over Mythos access for British banks

The UK is in talks with Anthropic to secure sector-wide Mythos access for British lenders, the Financial Times reported Friday. British banks are seeking guidance from U.S. counterparts already testing the model. The negotiations come after the Bundesbank's call for a "level playing field" for European banks, as reported by Bloomberg in AIPD's April 21st edition. The Financial Times did not name the UK agency leading the negotiations.

Read at FT ↗

OpenAI ships GPT-5.5 with agentic coding and computer-use emphasis

OpenAI launched GPT-5.5 Thursday with new capabilities for agentic coding, computer use and scientific research, the New York Times reported. The model is available to Plus, Pro, Business and Enterprise tiers and includes a new GPT-5.5 Pro tier. NYT framed the release as "a more open approach to cybersecurity than its chief rival, Anthropic," distinguishing OpenAI's posture from Anthropic's restricted Mythos rollout. Sam Altman published an accompanying post outlining iterative deployment safety principles.

Read at NYT ↗

II.China Watch

AmCham China flags Beijing security regime as top U.S. business concern before Trump summit

The American Chamber of Commerce in China released its 2026 white paper Thursday, citing broad and evolving data security definitions and complex national security reviews as continuing drags on U.S. firms' optimism, per SCMP. AmCham members also flagged nontariff barriers including rare earth export controls and licensing delays. Still, 79% of respondents held a positive or neutral outlook on U.S.-China relations in 2026, up 30 points from last year, and half ranked China among their top three investment destinations. Trump is scheduled to visit Beijing on May 14-15.

Read at SCMP ↗

Beijing issues new carbon guidelines targeting AI data centers

The Central Committee of the Chinese Communist Party and the State Council jointly issued an 18-point document directing energy intensive sectors including data centers to boost efficiency and accelerate decarbonization to meet Beijing's 2030 carbon peak pledge, SCMP reported. The guidelines call for cooling system upgrades and better energy efficiency per unit of computing power. The National Development and Reform Commission said the measures bolster energy security amid volatile prices from the U.S.-Israeli war in Iran. Local governments are required to file carbon action plans under the new regime.

Read at SCMP ↗

Alibaba Cloud expands China IaaS lead to 32.8% on AI demand

Alibaba Cloud's share of China's infrastructure-as-a-service market grew to 32.8% in 2025 from 30.1% in 2024, a 2.7-point gain driven by AI cloud demand, PingWest reported, citing Gartner's 2025 global IaaS market data. TechNode Global reported Alibaba also ranked first in Asia-Pacific IaaS revenue with 22.5% share, up from 20.8% a year earlier. Both outlets relied on Gartner's 2025 global and Asia-Pacific IaaS revenue rankings. U.S. hyperscalers AWS, Microsoft and Google remain constrained in mainland China by data localization and licensing rules.

Read at PingWest ↗

DeepSeek previews V4 model on the anniversary of the R1 release

Chinese AI lab DeepSeek released preview versions of its V4 model Friday, describing it as the most powerful open source AI platform, CNBC reported. The release lands roughly one year after V3/R1 surprised Silicon Valley with frontier-comparable Chinese capability. Bloomberg reported the preview omits published benchmark scores and formal license terms. DeepSeek did not announce a public release date for the full model.

Read at CNBC ↗

III.Federal Policy Tracker

CSET releases Director Helen Toner's written recommendations on frontier AI transparency

The Center for Security and Emerging Technology published Director Helen Toner's full written testimony Wednesday, expanding on her appearance at the Senate Judiciary Committee's "Stealth Stealing" hearing on China's theft of U.S. innovation, included in AIPD's April 22nd edition. CSET said the recommendations cover U.S.-China competition, frontier AI transparency and intellectual property protection. The written statement was posted to the CSET website. The release does not include specific draft legislative text.

Read at CSET ↗

GSA names new cohort of Presidential Innovation Fellows

The General Services Administration announced a new Presidential Innovation Fellows cohort Wednesday, FedScoop reported. The program brings technologists into government on fixed tours of duty. The cohort continues a program that has historically placed technologists at agencies running AI modernization efforts. GSA did not publish the cohort's size or specific agency placements.

Read at FedScoop ↗

IV.Industry & Market Watch

Cohere and Aleph Alpha announce $20 billion transatlantic AI tie-up framed around sovereign AI

Canada's Cohere and Germany's Aleph Alpha announced a $20 billion transatlantic AI tie-up Friday explicitly framed around providing "sovereign" AI systems independent of U.S. and China, the Financial Times reported. The Financial Times did not detail the $20 billion structure, whether equity, compute commitments or joint venture. Cohere is headquartered in Toronto; Aleph Alpha is headquartered in Heidelberg. The announcement did not disclose a closing date.

Read at FT ↗

Meta to cut 10% of workforce, cancel 6,000 open roles to "offset" AI capex

Meta will cut roughly 10% of its workforce, around 8,000 employees, and cancel 6,000 open roles to "offset" Mark Zuckerberg's AI spending, the Financial Times reported. The company plans up to $135 billion in capital spending this year, much of it on AI infrastructure. The Guardian, NYT and SiliconANGLE carried concurrent coverage of the announcement. The geographic distribution of the cuts was not announced.

Read at FT ↗

CrowdStrike launches Project QuiltWorks coalition for AI-discovered vulnerabilities

CrowdStrike launched Project QuiltWorks Wednesday, an industry coalition aimed at helping enterprises find and fix software flaws surfaced by frontier AI models, SiliconANGLE reported. The coalition responds to AI vulnerability discovery capabilities like those in Anthropic's restricted Mythos model. CrowdStrike did not publish a founding member list or governance structure. SiliconANGLE reported the launch was structured as an industry coalition rather than a CrowdStrike product.

Read at SiliconANGLE ↗

Thiel-backed Stark expands into defensive drones amid Iran war demand

Thiel-backed defense startup Stark expanded into defensive drones, driven by Iran war demand for UAV protection, the Financial Times reported. The company had previously focused on offensive drone systems. FT did not name the specific U.S. or allied contracts associated with the expansion. Stark did not publish a capacity target for the new product line.

Read at FT ↗

Tesla to upgrade customers whose cars cannot deliver Full Self-Driving

Tesla promised to upgrade customers who bought cars that cannot drive autonomously, the Wall Street Journal reported. The remediation acknowledges that some Tesla hardware cannot run the company's promised Full Self-Driving software. WSJ did not specify the size of the affected fleet. Tesla did not announce a cost estimate for the upgrade program.

Read at WSJ ↗

V.Global & Geopolitics

Microsoft commits A$25 billion ($18 billion) to Australia AI and cybersecurity buildout

Microsoft announced an A$25 billion ($18 billion) Australia investment Wednesday covering digital infrastructure, cybersecurity and AI development, CNBC reported. The investment is structured as a multiyear package. Microsoft did not publish a breakdown between data center and workforce skilling components. The Australian government has not detailed tax incentives associated with the announcement.

Read at CNBC ↗