SUBJECT: House targets Chinese AI model theft | a16z's $50M AI war chest | Mythos reaches UK banks

Thursday, April 16, 2026 — AI Policy Daily

- House Republicans draft **sanctions targeting Chinese AI model distillation**, extending the tech competition fight beyond chip export controls.

- Anthropic confirms **UK bank access to Mythos** within days, the cybersecurity model's first international deployment.

- Anthropic's own red team found **Mythos can compromise foundational computing systems**, fueling restricted-release decisions and government briefings.

- a16z founders pump $25M into an AI super PAC, pushing the **AI industry's midterm election war chest** past $50M.

- Apple and Google still host **nonconsensual sexualized image apps** despite explicit policies banning them, a new transparency report finds.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

1/ AI POLICY TODAY

• **House Republicans prepare sanctions targeting Chinese AI model distillation** — House Republicans are developing legislation to impose sanctions on Chinese entities that copy US AI models through distillation, Bloomberg reported. The measure targets the practice of training smaller models to replicate frontier US systems' outputs, moving beyond traditional intellectual property theft into AI-specific enforcement. The initiative is part of a broader congressional campaign to counter China in the global AI race alongside existing semiconductor export controls. Bill text and sponsors have not yet been publicly released. [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-16/us-house-to-weigh-penalties-on-ai-model-copying-by-chinese-firms)

• **Anthropic confirms Mythos access for UK banks within days** — Pip White, Anthropic's head of UK, Ireland, and northern Europe, confirmed that UK banks will receive access to the Mythos cybersecurity model "next week" in a Bloomberg interview. UK access would be the first international Mythos deployment beyond its initial US rollout to Silicon Valley partners and Wall Street banks. US banks began internal Mythos testing at government urging, as reported by Bloomberg in AIPD's [Apr 11 edition](https://aipolicydaily.org/archive/daily/2026-04-11/). White did not disclose which UK banks will participate or the terms of access. [Bloomberg](https://www.bloomberg.com/news/videos/2026-04-16/anthropic-s-mythos-available-to-uk-banks-in-next-week-video)

• **Anthropic's red team found Mythos could compromise foundational computing systems** — Anthropic's own researchers determined that Mythos could hack the systems underlying most modern computing infrastructure, according to a Bloomberg investigative feature published today. The report provides previously unreported detail on the internal red-teaming process and decision-making that led Anthropic to restrict Mythos's release. Banks and government agencies are assessing the threat posed by the model's demonstrated capabilities. Anthropic briefed the Trump administration on Mythos, as reported by TechCrunch in AIPD's [Apr 15 edition](https://aipolicydaily.org/archive/daily/2026-04-15/). [Bloomberg](https://www.bloomberg.com/news/features/2026-04-16/how-anthropic-discovered-mythos-ai-was-too-dangerous-for-release)

• **Andreessen and Horowitz inject $25M into AI super PAC, pushing total past $50M** — Marc Andreessen and Ben Horowitz poured $25 million into a pro-AI super political action committee, boosting the industry's political war chest to more than $50 million ahead of November's midterm elections, Bloomberg reported. The PAC backs candidates and positions favorable to AI industry priorities on regulation. AI policy is expected to be contested across multiple midterm races as public sentiment toward AI has turned negative. [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-15/andreessen-horowitz-boost-ai-super-pac-cash-to-over-50-million)

• **Apple and Google host nudify apps despite policies banning them, Tech Transparency Project finds** — Apple and Google continue offering mobile apps that generate nonconsensual sexualized images despite their own content policies, according to a Tech Transparency Project report published Wednesday, as reported by Bloomberg. The two largest app gatekeepers have failed to enforce their own rules against these apps. Apple separately threatened to remove Grok from its App Store over AI-generated nonconsensual images, as reported by NBC News in AIPD's [Apr 15 edition](https://aipolicydaily.org/archive/daily/2026-04-15/). [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-15/apple-google-offer-nudify-apps-despite-policies-against-them)

• **Doctors and education experts call for five-year moratorium on AI in schools** — A group of doctors and education researchers who studied AI's impact on young people are calling for a five-year moratorium on AI use in schools, Fortune reported in an exclusive. The researchers concluded that AI tools are being deployed in classrooms without adequate evidence of safety or efficacy. "If a local children's hospital told parents, 'We've got this new drug, it has potential, just trust us,' people would be horrified," one researcher told Fortune. Separately, a Wired and Indicator analysis this week underscored the urgency, identifying nearly 90 schools and 600 students worldwide affected by AI-generated deepfake nude images. [Fortune](https://fortune.com/2026/04/16/doctors-experts-ai-moratorium-schools/)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

2/ CHINA WATCH

• **China doubles domestic AI computing cluster to 60,000 chips using no US hardware** — China's largest AI computing cluster for scientific research began full operation in Zhengzhou this week, doubling from 30,000 to 60,000 domestically produced AI accelerator chips in two months, CCTV reported (state media). The chips were produced by Sugon, a Chinese Academy of Sciences affiliate that has been on the US Commerce Department Entity List since 2019. The cluster powers the core node of China's national supercomputing network. Sugon's chips are entirely domestically manufactured with no US components. [SCMP](https://www.scmp.com/news/china/science/article/3350208/china-doubles-ai-science-computing-scale-2-months-using-no-us-chips)

• **China tests deep-sea cable cutter at 3,500-meter depth** — China's "Haiyang Dizhi 2" research vessel successfully tested a deep-sea electro-hydrostatic actuator capable of cutting submarine cables at 3,500 meters, per the Ministry of Natural Resources (govt agency). The self-contained tool can be mounted on unmanned underwater vehicles, and state media described the trial as bridging the "last mile" from development to deployment readiness. Approximately 95% of intercontinental data flows through undersea cables at similar depths. [SCMP](https://www.scmp.com/news/china/science/article/3350174/china-tests-submarine-cable-cutter-3500-metre-depth)

• **Chinese AI startup StepFun restructures for Hong Kong IPO at up to $10B valuation** — StepFun, a Shanghai-based foundation model developer backed by state-linked investors and Tencent, is unwinding its offshore corporate structure to pursue a Hong Kong IPO at a valuation of up to $10 billion for anchor investors, per Reuters. The company was founded by former Microsoft executive Jiang Daxin. The restructuring follows Beijing's tightening scrutiny of offshore-registered tech firms. StepFun plans to file by June. [Reuters via The Standard](https://www.thestandard.com.hk/editorial/article/329151/Chinese-AI-startup-StepFun-to-unwind-offshore-structure-to-pave-way-for-IPO-sources-say)

• **JD.com launches robot repair service, plans expansion to 50+ Chinese cities** — JD.com launched a service offering repair, diagnostics, battery replacement, and recycling for humanoid robots, quadruped robots, and AI companion devices, per TechNode. The service launched in Beijing and will expand to more than 50 cities within three years. Chinese firms Unitree and Agibot accounted for the large majority of global humanoid robot shipments in 2025, with combined output exceeding 10,000 units, creating a growing installed base requiring aftermarket support. JD.com's service covers products from multiple Chinese robotics manufacturers. [TechNode](https://technode.com/2026/04/16/jd-com-launches-robot-ambulance-service-plans-expansion-to-50-cities-across-china/)

• **Alibaba tops international robotics benchmark for second time in two weeks** — Alibaba's ABot-PhysWorld, a world model for robotic manipulation, topped the WorldArena benchmark developed by Tsinghua University and partner institutions. It surpassed Google Veo and other leading models in physical plausibility and action prediction. Alibaba has now won two major benchmarks in under two weeks; its HappyHorse model topped the Artificial Analysis leaderboard in early April. The wins span embodied AI and general-purpose model performance. [QbitAI](https://www.qbitai.com/2026/04/401554.html)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

3/ FEDERAL POLICY TRACKER

**Executive Branch**

• **DOE launches AI testbed to evaluate models for energy sector operations** — The Department of Energy's Office of Cybersecurity, Energy Security and Emergency Response partnered with Lawrence Livermore National Laboratory to launch a dedicated AI evaluation platform for energy infrastructure, FedScoop reported. The testbed allows utilities, grid operators, vendors, national labs, and research organizations to upload AI models for vulnerability and risk assessment. The platform is now operational and accepting submissions. [FedScoop](https://fedscoop.com/doe-ai-testbed-cybersecurity-lawrence-livermore-energy/)

**State Activity**

• **Pennsylvania expands generative AI to 3,000 state employees, thousands more in training** — Pennsylvania expanded generative AI tools to 3,000 state government employees, with thousands more currently in training, StateScoop reported. The expansion fulfills a 2023 governor's executive order on AI adoption in state government. The deployment is among the largest state-level generative AI rollouts in the country. [StateScoop](https://statescoop.com/pennsylvania-expands-generative-ai-tools-3000-employees/)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

4/ INDUSTRY & MARKET WATCH

• **ASML shares fall 5% despite raising guidance as China export restrictions tighten** — ASML, the sole supplier of extreme ultraviolet lithography machines used to manufacture advanced AI chips, saw its stock drop 5% despite beating Q1 2026 earnings expectations and raising its full-year guidance, CNBC reported. The decline was driven by tightening export restrictions on sales to China, which is limiting ASML's addressable market. The stock reaction demonstrates the direct economic cost to allied companies of US-led semiconductor export controls. ASML raised its 2026 revenue outlook despite the export restriction headwinds. [CNBC](https://www.cnbc.com/2026/04/15/asml-q1-2026-earnings-report.html)

• **Public opinion on AI and data centers sours as OpenAI and Anthropic eye IPOs** — Negative public sentiment toward AI and data centers is growing, with potential effects on IPO timing for OpenAI and Anthropic and the politics of AI regulation heading into November, CNBC reported. Opposition to data centers has intensified, with Maine banning new construction and Sen. Sanders introducing a federal moratorium in recent weeks. The public backlash coincides with midterm elections where AI regulation is expected to be contested across multiple races. Tech companies continue expanding AI infrastructure spending despite the opposition. [CNBC](https://www.cnbc.com/2026/04/15/public-opinion-ai-data-centers-anthropic-openai-ipo.html)

• **OpenAI steps back from Stargate Norway as Microsoft takes over the project** — OpenAI is pulling back from its planned Stargate data center project in Norway, with Microsoft taking over the development, CNBC reported. OpenAI is now in discussions with Microsoft about renting compute capacity from the facility instead of operating it directly. The restructuring extends Microsoft's de facto control over AI infrastructure assets as OpenAI prepares for an IPO emphasizing its independence. OpenAI would become a tenant in the Microsoft-operated Norway facility. [CNBC](https://www.cnbc.com/2026/04/15/openai-stargate-norway-project-microsoft.html)

• **AI-driven 'right sizing' template spreads from Snap to Block to Amazon** — A pattern of AI-justified workforce restructuring is spreading through major corporations, with Snap, Block, and Amazon adopting similar approaches, the Wall Street Journal reported. Snap cut roughly 1,000 employees this week, explicitly citing AI capabilities as the rationale. Block and Amazon have pursued parallel restructuring strategies that frame layoffs as technology-enabled efficiency gains. Widespread AI-driven workforce reduction could prompt congressional action on AI displacement. Multiple other companies are evaluating similar approaches, according to the WSJ. [WSJ](https://www.wsj.com/business/has-the-era-of-the-mega-layoff-arrived-928f061d?mod=pls_whats_news_us_business_f)

• **LinkedIn data shows hiring down 20% since 2022 but blames interest rates, not AI** — LinkedIn's analysis of its platform data shows hiring is down approximately 20% since 2022 but attributes the decline to higher interest rates rather than AI displacement, TechCrunch reported. The data provides a counter-narrative to the AI-driven layoff fears highlighted by this week's corporate restructuring announcements. Whether the hiring decline is rate-driven or AI-driven has direct implications for the design of workforce policy responses. LinkedIn's platform covers more than one billion users, making it one of the largest labor market datasets available. [TechCrunch](https://techcrunch.com/2026/04/15/linkedin-data-shows-ai-isnt-to-blame-for-hiring-decline-yet/)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

5/ GLOBAL & GEOPOLITICS

• **EBA head warns European banks to prepare for AI-driven cybersecurity threats** — European Banking Authority head François-Louis Michaud warned European banks to prepare for AI-driven cybersecurity threats, saying lenders are resilient to current shocks but must brace for future risks, Reuters reported. The warning came the same day Anthropic confirmed UK bank access to its Mythos cybersecurity model. The EBA's statement addressed AI cybersecurity threats broadly rather than referencing any specific model. Michaud said European lenders must strengthen their defenses against AI-enabled attack vectors. [Reuters](https://www.reuters.com/sustainability/boards-policy-regulation/european-banks-can-withstand-current-shocks-watchdog-head-says-2026-04-16/)

• **Nvidia CEO Huang says Mythos demonstrates need for US-China AI safety cooperation** — Nvidia CEO Jensen Huang argued that Anthropic's Mythos cybersecurity capabilities demonstrate the need for greater US-China cooperation on AI safety, Bloomberg reported. Huang said researchers in both economies must agree on frameworks for safely managing increasingly powerful AI models. Huang's call comes as Nvidia faces ongoing US export controls that restrict chip sales to China. Nvidia's advanced AI chips remain central to both the US capability advantage and the export control regime. [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-15/nvidia-s-huang-says-mythos-shows-need-for-us-china-ai-dialogue)