---
date: 2026-05-06
subject: "CAISI signs Google/MSFT/xAI testing pact | PA sues Character.AI | Apple pays $250M Siri"
---

**Commerce's CAISI inked pre-release testing deals** with Google DeepMind, Microsoft and xAI, granting government evaluators access to frontier models with safeguards stripped to probe national security risks. **Pennsylvania filed suit against Character.AI** in Commonwealth Court, alleging its chatbots impersonated licensed psychiatrists and cited a bogus state license number to an investigator. **Apple will pay $250 million** to settle class claims that it marketed Siri capabilities tied to Apple Intelligence that the company later postponed.

# 1. AI Policy Today

- **Commerce's CAISI signs binding pre-deployment testing agreements with Google DeepMind, Microsoft and xAI** — The Center for AI Standards and Innovation, the AI evaluation unit at NIST, said Tuesday it has signed agreements with Google DeepMind, Microsoft and xAI to evaluate frontier models before public release. The agreements give CAISI access to models with reduced or removed safeguards to evaluate national security capabilities and risks, per WSJ. CAISI's prior partnerships with OpenAI and Anthropic, first launched in 2024, have been renegotiated to reflect Trump AI Action Plan priorities, per Axios. The action operationalizes the pre-release vetting framework the New York Times reported the White House was weighing, as flagged in AIPD's [May 5th edition](https://aipolicydaily.org/archive/daily/2026-05-05/). Anthropic, which the Pentagon designated a supply chain risk, is not among Tuesday's three named signatories. [NIST](https://www.nist.gov/news-events/news/2026/05/caisi-signs-agreements-regarding-frontier-ai-national-security-testing) [WSJ](https://www.wsj.com/tech/ai/google-microsoft-and-xai-agree-to-share-early-ai-models-with-u-s-f95a88d1?mod=rss_Technology) [Axios](https://www.axios.com/2026/05/05/us-frontier-ai-testing-white-house-pivots-safety)

- **Pennsylvania Gov. Shapiro sues Character.AI for chatbots posing as licensed psychiatrists** — Pennsylvania Gov. Josh Shapiro and the state Department of State announced Tuesday that they filed suit in Commonwealth Court against Character Technologies, alleging Character.AI chatbots impersonated licensed medical professionals, The Hill reported. The complaint says one chatbot falsely identified itself as a Pennsylvania-licensed psychiatrist and provided an invalid state license number during an investigation by the agency's AI Task Force. CBS News reported one Character.AI chatbot named Emilie claimed it could assess and prescribe within "my remit as a Doctor." Shapiro's office described the action as a first-of-its-kind enforcement matter, per AP News. The suit asks the court to stop the company's chatbots from engaging in the unlawful practice of medicine and surgery, and seeks a preliminary injunction under Pennsylvania's Medical Practice Act. [The Hill](https://thehill.com/policy/healthcare/5864427-pennsylvania-lawsuit-ai-chatbots-doctors-therapists/) [AP](https://apnews.com/article/46502067ed5b3cd9f9173f194ad30070) [CBS](https://www.cbsnews.com/amp/news/pennsylvania-character-ai-lawsuit-chatbot-posed-as-medical-professional/)

- **Anthropic's Amodei warns of 6-to-12-month AI cyber "moment of danger"** — Anthropic Chief Executive Dario Amodei warned at a JPMorgan event with the bank's Chief Executive Jamie Dimon on Tuesday that AI has created a 6-to-12-month window for software firms, governments and banks to fix tens of thousands of vulnerabilities being uncovered by tools such as Anthropic's Mythos model, CNBC reported. Amodei said the threat is "some enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that's done from ransomware on schools, hospitals, not to mention banks," per Decrypt. The remarks came at a nearly two-hour event tied to Anthropic's push into financial services, where the company unveiled Claude-based agents for pitchbooks, earnings review and compliance work. Earlier testing of Mythos with Mozilla had identified 271 vulnerabilities. [CNBC](https://www.cnbc.com/2026/05/05/anthropic-ceo-cyber-moment-of-danger-mythos-vulnerabilities.html) [Decrypt](https://decrypt.co/366891/anthropic-warns-cyber-risk-window-ai-uncovers-flaws/)

- **Apple agrees to $250 million class settlement over delayed Apple Intelligence Siri features** — Apple agreed to pay $250 million to settle a class-action suit alleging it misled iPhone buyers by promoting Apple Intelligence Siri features that the company later delayed, the New York Times reported. Settlement payments will go to U.S. class members who submit claim forms, with $25 per eligible device and an increase up to $95 per device if claim volume is low, per MacRumors. Eligible devices include iPhone 16 and 16 Pro variants and iPhone 15 Pro and 15 Pro Max purchased between June 10, 2024 and March 29, 2025. Apple did not admit wrongdoing and said it has shipped multiple Apple Intelligence features since 2024 including Visual Intelligence, Live Translation and Writing Tools. The settlement received preliminary approval and class members will start receiving notices within 45 days. [NYT](https://www.nytimes.com/2026/05/05/technology/apple-intelligence-lawsuit-settlement.html) [MacRumors](https://www.macrumors.com/2026/05/05/apple-class-action-siri-lawsuit-settlement/)

- **Five publishers and Scott Turow sue Meta over Llama, allege "verbatim" copying of textbooks and novels** — Hachette, Macmillan, McGraw Hill, Cengage Learning and Elsevier, joined by author Scott Turow, filed a proposed class-action complaint Tuesday in Manhattan federal court alleging Meta pirated millions of copyrighted works to train its Llama AI models without permission, the Guardian reported. The plaintiffs allege Llama generates verbatim reproductions of original passages and mirrors authors' personal styles in its outputs. The complaint accuses Meta of scraping pirate sites and of removing copyright management information from the works to conceal the practice. The suit names Mark Zuckerberg as a defendant, alleging he personally authorized the use of pirated collections to train Llama. The publishers seek damages and a jury trial. [Guardian](https://www.theguardian.com/technology/2026/may/05/publishers-sue-meta-copyright-ai)

- **U.S. Army opens "Right to Integrate" hackathons, asking missile primes to red-team weapons software** — The U.S. Army is asking missile makers and other defense contractors to open up weapons software to AI driven testing tools through a new program called "Right to Integrate," WSJ reported. The first one-day session will run later this month at Fort Carson, Colorado, with Anduril, Boeing, General Dynamics, L3Harris, Leidos, Lockheed Martin, Northrop Grumman, Palantir, Perennial Autonomy and RTX taking part, per Defense One. Army Secretary Dan Driscoll said the war in Ukraine "showed the world that speed matters and an open architecture construct is highly effective in high-intensity warfare. We haven't been moving fast enough." Army Chief Technology Officer Alex Miller said existing systems remain held back by poor standards adoption, closed proprietary interfaces and an inability to adapt over time. The hackathons will run through a series of sessions to integrate platforms with the service's next-generation command-and-control software. [WSJ](https://www.wsj.com/tech/army-asks-missile-makers-to-hack-their-own-weapons-a6bf3288?mod=rss_Technology) [Defense One](https://www.defenseone.com/defense-systems/2026/05/army-turns-hackathons-connect-dozens-battlefield-and-business-systems/413335/)

- **Meta deploys AI age estimation using height and bone structure to flag underage Facebook and Instagram users** — Meta said Tuesday it will scan photos and videos for visual cues such as a person's height or bone structure to estimate whether a user is under 13, TechCrunch reported. The company said in a blog post the system is not facial recognition and looks at general themes and visual cues to estimate general age rather than identify a specific person. The system is operating in select countries, with Meta saying it will broaden the rollout and extend the technology to Instagram Live and Facebook Groups. Meta will deactivate accounts it determines are underage, with users required to complete the company's age verification process to avoid deletion. The announcement comes weeks after a New Mexico jury ordered Meta to pay $375 million in civil penalties for misleading consumers about platform safety for children. [TechCrunch](https://techcrunch.com/2026/05/05/meta-will-use-ai-to-analyze-height-and-bone-structure-to-identify-if-users-are-underage/)

# 2. China Watch

- **China unveils fully domestic 2-exaflop supercomputer designed to bypass U.S. chip controls** — China has unveiled a new supercomputer built from homegrown CPUs that is designed to reach 2 exaflops, edging past the 1.8-exaflop El Capitan at Lawrence Livermore National Laboratory, the South China Morning Post reported. Known as Lingsheng or LineShine, the system runs entirely on CPUs rather than the GPUs that power other exascale machines. The article said the design is intended to sidestep U.S. export controls on advanced AI accelerators. [South China Morning Post](https://www.scmp.com/news/china/science/article/3352514/china-targets-top-spot-supercomputing-fully-domestic-machine?utm_source=rss_feed)

- **ChinaTalk maps gray-market "transfer stations" reselling Claude tokens at deep discounts inside China** — "Transfer station" operators in China resell access to Anthropic's Claude and other U.S. AI models at roughly 1 yuan per dollar of tokens, a 70% to 90% discount to official prices, ChinaTalk reported. Operators bulk-register developer accounts, recruit verification participants in lower-income countries, aggregate user requests to lower per-call cost, and route Alipay or WeChat payments through software interfaces that proxy traffic to Anthropic. The piece said model providers can suspend individual operators, but upstream account pools and downstream customer bases stay intact, letting replacement stations stand up quickly. The structure routes paid Chinese demand to U.S. labs while bypassing identity verification and geographic safeguards. It undermines a key tool the Commerce Department and Treasury rely on to track AI exposure to restricted Chinese end users. [ChinaTalk](https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens-in)

- **China's Big Fund in talks to lead DeepSeek round at $45 billion valuation** — China's main chip sector investment vehicle, the China Integrated Circuit Industry Investment Fund, is in talks to lead DeepSeek's first outside fundraising at a valuation of about $45 billion, the Financial Times reported, as carried by Bloomberg. The state-backed Big Fund is also a backer of Semiconductor Manufacturing International Corp. Details of the round and its participants have not been finalized. The talks come days after the Commerce Department's CAISI published an evaluation of DeepSeek models. [Bloomberg](https://www.bloomberg.com/news/articles/2026-05-06/china-chip-fund-in-talks-to-lead-mega-deepseek-funding-ft-says)

- **DeepSeek pulls its first multimodal vision paper hours after release** — DeepSeek published its first technical paper describing visual capabilities and then deleted it overnight, PingWest reported. The paper marked the lab's first publicly disclosed multimodal effort, building on the limited vision mode testing covered in AIPD's [May 1st edition](https://aipolicydaily.org/archive/daily/2026-05-01/). The deletion came within hours of the paper's release. The release-and-retract pattern leaves U.S. agencies with limited insight into Chinese frontier capability disclosures. It undermines the structured disclosure framework that the Commerce Department and CAISI rely on to assess foreign AI risk. [PingWest](https://www.pingwest.com/a/313460)

# 3. Federal Policy Tracker

- **[Reg] FAA proposes rule to designate critical infrastructure drone-free zones** — The Federal Aviation Administration on Tuesday proposed rules letting owners of facilities such as energy plants, oil refineries, state prisons and amusement parks request restrictions on drone flights near their sites, the New York Times reported. Applicants must justify a restriction on aviation safety, ground protection, national security or homeland security grounds and describe drone operations over the facility during the previous 24 months. The proposal implements a 2016 congressional directive and a 2025 Trump executive order, per Reuters. The notice arrives as the U.S. prepares to host the 2026 FIFA World Cup and 2028 Summer Olympics, with the FAA already probing operators it says flew drones too close to Major League Baseball games at Coors Field in Denver. The rules will be published in the Federal Register on May 6. [NYT](https://www.nytimes.com/2026/05/05/us/politics/faa-drones-proposed-rule.html)

- **[Reg] CMS official says AI gives agency "longer leash" to pursue claims fraud** — Jeneen Iwugo, acting director of the CMS Center for Program Integrity, said the Center has shifted away from a previously conservative approach to fraud pursuit under the Trump administration, FedScoop reported. Iwugo said AI flags suspicious activity daily against the 4 to 5 million Medicare and Medicaid claims the agency processes. She said the new posture gives the Center "a longer leash" to pursue cases. The Center for Program Integrity oversees fraud and abuse work for Medicare and Medicaid, the largest U.S. health payment programs. [FedScoop](https://fedscoop.com/cms-ai-fraud-detection-medicare-medicaid-programs/)

# 4. Industry & Market Watch

- **Financial Stability Board warns AI driven private credit boom risks "sizeable" losses on data center oversupply** — The Financial Stability Board, the global watchdog that monitors central banks in 24 countries, said in a new report that AI loans dominate private credit and that a sharp correction in asset valuations could lead to sizeable credit losses to private credit investors, the Guardian reported. The AI industry accounted for more than a third of private credit deals in 2025, up from 17% over the previous five years. The FSB said triggers could include any major shortfall in electricity supply for data center construction, leading to delays or project cancellations, or an oversupply of data centers outpacing AI demand. The report flags healthcare, services and tech as the largest borrowers from private credit funds, which lend outside the regulated banking system using investor money. The warning lands as private credit funds have already seen multibillion-pound withdrawals that forced some to cap client redemptions. [Guardian](https://www.theguardian.com/business/2026/may/06/global-finance-watchdog-warns-over-private-credit-industry-fuelling-ai-boom)

- **DeepMind UK staff cite Pentagon Gemini deal and Iran war in union vote, Wired reports** — Workers at Google DeepMind in London voted to unionize as part of an effort to block the lab's AI from being used in U.S. and Israeli military settings, Wired reported, building on the Guardian's exclusive from AIPD's [May 5th edition](https://aipolicydaily.org/archive/daily/2026-05-05/). Communication Workers Union national officer for technology John Chadfield told Wired: "Fundamentally, the push for unionization is about holding Google to its own ethical standards on AI, how they monetize it, what the products do, and who they work with." A DeepMind employee told the magazine that Google's Pentagon clause permitting any lawful government use of its AI is "vague enough to be effectively meaningless." Google said in a statement that no formal unionization vote has yet occurred and pointed to its policy of constructive dialogue with employees. Roughly 600 U.S.-based Google employees signed a separate letter protesting the Pentagon Gemini deal. [Wired](https://www.wired.com/story/google-deepmind-workers-vote-to-unionize-over-military-ai-deals/)

- **Google pushes Gemini agentic workforce pitch to U.S. federal agencies** — Google is positioning Gemini for federal agentic workforce deployments and marketing AI agents to public-sector buyers as an immediate force multiplier rather than a future capability, SiliconANGLE reported, citing theCUBE Research. The piece said government agencies are becoming a proving ground for large-scale AI workforce transformation amid mounting operational demands, aging application environments and strict compliance requirements. The marketing arrives a week after the Pentagon confirmed signed Gemini contracts for classified network use under any lawful government purpose terms. The federal pitch broadens Google's public-sector push beyond its existing Pentagon agreements. [SiliconANGLE](https://siliconangle.com/2026/05/05/agentic-workforce-government-ai-gemini-thecuberesearch/)

# 5. Global & Geopolitics

- **EU and Japan agree on new AI, quantum and chips cooperation steps at Brussels Digital Partnership Council** — The European Union and Japan agreed to deepen regulatory, research and industry cooperation on AI, data, quantum, semiconductors, digital infrastructure and online platforms at the fourth meeting of the EU-Japan Digital Partnership Council in Brussels, the European Commission announced. The new steps will improve cross-border data flows, advance interoperable digital identities and strengthen cooperation on platform regulation. The meeting was led by EU Executive Vice-President for Tech Sovereignty Henna Virkkunen with Japanese Digital Transformation Minister Hisashi Matsumoto, Internal Affairs Minister Yoshimasa Hayashi and METI Parliamentary Vice-Minister Toshiyuki Ochi. The Council framework operates under the EU's Digital Services Act cooperation track. [European Commission](https://digital-strategy.ec.europa.eu/en/news/eu-and-japan-accelerate-cooperation-ai-data-quantum-and-chips)

- **NHS orders hundreds of GitHub repos closed by May 11, citing Mythos class AI risk** — The UK's National Health Service is ordering all of its technology leaders to switch hundreds of public GitHub repositories to private by May 11, citing Anthropic's Mythos and other advanced AI models capable of "large-scale code ingestion, inference, and reasoning," The Register reported, citing internal guidance. The decision was approved by the NHS Engineering Board. NHS England told The Register the move is a temporary measure while the organization assesses the impact of rapid developments in AI models. The order reverses a longstanding UK government policy that public-money software should default to open source. NHS sources said few of the affected repos contain anything sensitive, with most hosting documentation, architecture diagrams and codebases for internal scheduling tools. [The Register](https://go.theregister.com/feed/www.theregister.com/2026/05/05/nhs_to_closesource_hundreds_of_repos/)

- **India's SEBI issues infosec advisory citing Mythos by name, forms taskforce for the equities industry** — The Securities and Exchange Board of India on Tuesday advised participants in the country's equities industry to immediately revisit their information security systems and practices in case Anthropic's Mythos sparks a cyberattack spree, The Register reported. The advisory cites "Claude Mythos" by name and says AI driven vulnerability identification tools introduce heightened risk exposure and concerns over data confidentiality and application integrity. SEBI has established a taskforce to examine Mythos class risks, share threat intelligence, report incidents and review cybersecurity at third-party software vendors. The directive applies to 19 classes of regulated entity ranging from venture capitalists and merchant bankers to mutual funds and stock exchanges. The advisory follows U.S. Treasury Secretary Scott Bessent's emergency meeting with major banks weeks back and a similar advisory by Singaporean regulators on Tuesday. [The Register](https://go.theregister.com/feed/www.theregister.com/2026/05/06/india_seb_mythos_infosec_advice/)

- **Modi and To Lam agree to expand AI cooperation as Vietnam president visits New Delhi** — Indian Prime Minister Narendra Modi held bilateral talks with Vietnam President To Lam at Hyderabad House on Wednesday, Bloomberg reported. The two countries are looking to strengthen economic and defense ties as tensions persist in the Middle East. Lam met India's National Security Adviser Ajit Doval on May 5 and the two sides agreed to expand cooperation in security, AI, digital transformation, innovation and energy, per a Vietnamese government statement carried by The Straits Times. The visit is Lam's second overseas trip since securing the presidency in April. The two countries are seeking to deepen regional ties as relations with Washington remain unpredictable and as the Iran war disrupts trade with the Middle East. [Bloomberg](https://www.bloomberg.com/news/articles/2026-05-06/modi-lam-to-meet-as-india-and-vietnam-deepen-ai-defense-ties) [Straits Times](https://www.straitstimes.com/asia/south-asia/modi-lam-meet-as-india-and-vietnam-deepen-ai-defense-ties)