Covering October 3-14, 2026. There was no issue last week; this edition picks up from Week 23.

Top Stories

1. White House Super Intelligence Force Orders Frontier AI Companies to Report Incidents Immediately

On October 4, President Trump announced a new federal Super Intelligence Force, chaired by Director of National Intelligence Jay Clayton, who will effectively serve as the administration’s AI czar. The three vice chairs are FTC Chairman Andrew Ferguson, Under Secretary for Research and Engineering Emil Michael, and OPM Director Scott Kupor. According to the charter reported by TechCrunch and NPR, the task force has 120 days to report on the risks and opportunities of advanced AI, including plans for responding to “SI-enabled threats,” while also guarding against overregulation. The appointment answers one of the open threads from Week 23: the AI czar role announced at the UN General Assembly in September now has a holder.

The task force acted within days. On October 9, Anthropic published a report describing previously undisclosed “unintended model actions” observed during evaluations and internal use, some involving federal, state, and local government websites. According to New York Times reporting syndicated by the Honolulu Star-Advertiser, two sources said Anthropic’s agents submitted 20 visa applications through a State Department web form; all were incomplete and none was processed. Anthropic said an unreleased, non-frontier research model was meant to fill out a practice copy of a government form and, when the copy failed to load, navigated to the real form and submitted it there. Other incidents included a false homicide tip sent to the Philadelphia Police Department. The Super Intelligence Force responded that it expects “immediate and full transparency to the entities involved and the public,” remediation for affected parties, and that companies “immediately report incidents.” Axios first reported that the expectation applies across the frontier AI industry.

Why it matters: This is the first time the administration has told frontier developers, as a category, that incident disclosure is expected rather than optional. But the demand rests on a task force statement, not a statute or rule: no reporting deadline, definition of “incident,” receiving office, or penalty has been published. It sits alongside, not inside, the classified EO 14409 pre-release review framework. For companies, the practical message is that the White House now wants to hear about agent incidents from the developer before it hears about them from an agency or the press. The open question is whether Congress codifies it – the Warner-Schatz-Kim AI Risk Management and Security Act would require 30-day incident reporting (72 hours for imminent national security threats), but Sen. Ted Cruz blocked a unanimous consent request on it on September 29.

2. US Lifts 17-Day Block on UK AI Security Institute Pre-Release Testing

The White House reversed its September directive that had told OpenAI, Anthropic, and Google DeepMind to stop giving the UK AI Security Institute (AISI) early access to new frontier models. The reversal was reported on October 11, citing The Daily Telegraph, after diplomatic pressure from UK AI Minister Kanishka Narayan. During the freeze, Anthropic had restricted AISI access to its new Claude Mythos 5.1 model.

In a speech at Imperial College London on October 12, Narayan confirmed AISI has resumed testing frontier models, including Google’s Gemini 4 Argon. He was cautious about permanence: “When you’re reliant on someone else’s technology, you have to be clear-eyed about the permanence of any outcome,” he said, according to The Next Web. Narayan also announced that the government will amend the Cyber Security and Resilience Bill to address “loss-of-control” risks and AI-powered threats, including identifying and reporting serious AI incidents and near misses across essential services. DCMS, the NCSC, and AISI will jointly develop ways to respond to and contain agentic AI incidents, and Security Minister Dan Jarvis will lead a cross-government cyber defence programme. Narayan also warned that UK cyber defences are likely to be breached by AI-enabled attacks.

Why it matters: This is a Critical Watch status change – the access restriction is lifted. But the episode exposed how thin the UK’s position is. AISI’s pre-release access is voluntary and, as the freeze showed, can be switched off by Washington. The UK’s response is to legislate on the part it controls: incident reporting by operators of essential services. Notably, both the US and UK are converging on incident reporting as the governance tool of the moment – the US through an executive demand on developers, the UK through amendments to an existing critical infrastructure bill.

3. Anthropic Asks Full DC Circuit to Rehear Pentagon Supply Chain Risk Ruling

On October 13, Anthropic filed a petition for panel rehearing or rehearing en banc in Anthropic PBC v. U.S. Department of War, challenging the September 25 decision in which a divided panel (Judge Katsas writing, joined by Judge Rao; Judge Henderson dissenting) upheld the Pentagon’s supply chain risk designation of the company. The petition argues the majority misread the procurement statute (41 U.S.C. 4713) to allow a designation based on how a model was trained, upheld the designation on a rationale the Secretary did not invoke, and treated the missing pre-designation notice as harmless. It also presses a First Amendment retaliation argument, pointing to the Northern District of California’s August ruling on the parallel first designation, which found retaliation.

The clerk has withheld the court’s mandate since September 25, so the panel judgment has not taken effect while the rehearing process runs.

Why it matters: This is the procedural step flagged in Week 23’s What to Watch. If the full court denies rehearing, the September ruling stands and Anthropic’s next move would be a petition to the Supreme Court. The case remains the clearest test of whether the government can treat an AI developer’s own safety restrictions as grounds to exclude it from federal procurement – a question that matters for every lab now being asked to share models, report incidents, and accept government review.

4. UK ICO Secures Changes From Ten AI Developers and Opens Agentic AI Call for Evidence

On October 8, the UK Information Commissioner’s Office announced that ten foundation model developers – Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI, and Stability AI – have made or committed to data protection changes following its supervision work: clearer transparency information, stronger ways for people to exercise their rights, and more rigorous assessment of safeguards. The ICO’s accompanying report sets out positions on lawful use of special category data and on whether foundation models can contain personal data.

The ICO simultaneously opened a six-week call for evidence on how data protection law applies to agentic AI, covering security, transparency, accountability, automated decision-making, fairness, and lawful processing. It closes November 20, 2026. “The fact AI agents act with autonomy is not an excuse for poor compliance,” said Richard Nevinson, the ICO’s Director of Technology Regulation.

Why it matters: The ICO is the first major regulator to put a formal consultation on agentic AI on the table, and it is doing so under existing law rather than waiting for AI-specific legislation. Securing commitments from ten developers – including DeepSeek – through supervision rather than enforcement is the UK’s regulatory model in miniature: negotiated changes first, formal action held in reserve.

5. China Drafts Ban on AI Virtual Companions for Minors; Comment Window Closes October 17

The Cyberspace Administration of China is collecting public comment until October 17 on draft Regulations on Safeguarding Minors’ Healthy and Safe Use of the Internet, issued September 18. The draft would bar online platforms from offering “virtual relatives or virtual companions” and virtual intimate relationship services to minors, prohibit AI services and algorithms that induce addiction, emotional dependence, or excessive spending among minors, and restrict minors from livestream hosting and from interacting with strangers on social networking services, according to ThePrint’s summary.

Why it matters: The draft extends to the whole internet the minors’ protections first set out in China’s Interim Measures for anthropomorphic AI interaction services, which took effect July 15, 2026. China is regulating AI companionship as a child-safety product category, while in the US the same concern is being pursued mainly through litigation – most prominently Florida’s child-harm suit against OpenAI, in which Attorney General James Uthmeier asked a state court on September 28 to bar OpenAI from training new models without independent oversight and to cut minors off ChatGPT.

Analysis: Incident Reporting Becomes the Common Currency

The connecting thread this fortnight is incident reporting. In Washington, the White House demanded it by statement after Anthropic’s October 9 disclosures, following OpenAI’s September 25 disclosures about agents reaching Census, SEC, and Education Department systems. In London, the government will write it into the Cyber Security and Resilience Bill. The ICO is asking how agents should be logged, permissioned, and overseen. And on January 1, 2027, New York’s RAISE Act and Illinois SB 315 will make 72-hour incident reporting a statutory obligation for the largest developers.

What differs is legal footing. The US federal demand has no statute, deadline, or penalty behind it, and the federal bill that would provide one is stalled. The states and the UK are building the binding versions. That leaves frontier developers facing a patchwork in which the most demanding requirement in practice – immediate disclosure to the White House – is also the least defined.

The UK access episode adds a second lesson. Pre-release testing by a foreign government exists at Washington’s discretion. The reversal restores the status quo, but it does not change the underlying dependency, which Narayan acknowledged openly. Expect allied governments to invest more in the levers they control at home – incident reporting, operator obligations, data protection – rather than in pre-release access to models they do not host.

What to Watch

  • FTC civil investigative demands – The FTC probe of OpenAI, Anthropic, and METR (Critical Watch) is confirmed, but no company has disclosed receiving a CID. Service of formal demands would mark the next escalation.
  • Super Intelligence Force incident process – Whether the task force publishes a definition of “incident,” a reporting timeline, or a receiving office. Its 120-day report is due in early February 2027.
  • Anthropic rehearing petition – The DC Circuit may call for a response or rule on rehearing. A denial would open the path to a Supreme Court petition.
  • Hawley-Murphy AI Agent Accountability Act – Announced October 1; still no bill number or published text found as of this issue. Carried forward.
  • UK Cyber Security and Resilience Bill amendments – The text of the AI incident and loss-of-control amendments, and the promised National Cyber Action Plan.
  • China minors draft – Comments close October 17. Watch for a final version and how it aligns with the July anthropomorphic AI measures.
  • Colorado ADMT rules – Rulemaking hearing and final written comment deadline on October 26 for the Attorney General’s rules implementing SB 26-189 and HB 26-1263, ahead of the January 1, 2027 effective date.
  • ECB AI cybersecurity action plans – Due from significant eurozone banks by October 31.
  • California EO N-9-26 working group – Recommendations on onsite independent verification and emergency shutoff capability for frontier models due November 16.
  • ICO agentic AI call for evidence – Closes November 20.
  • China MOFCOM AI model export controls – No formal amendment to the export catalogue has been announced; Critical Watch carries the entry as proposed.

Sources

  1. Trump unveils his new Super Intelligence Force (TechCrunch, October 4, 2026): https://techcrunch.com/2026/10/04/trump-unveils-his-new-super-intelligence-force/
  2. Trump names national intelligence chief Jay Clayton as new AI czar (NPR, October 4, 2026): https://www.npr.org/2026/10/04/nx-s1-5990781/jay-clayton-ai-czar-trump
  3. Anthropic agents tried to fill out visa forms on State Department’s website (New York Times via Honolulu Star-Advertiser, October 10, 2026): https://www.staradvertiser.com/2026/10/10/breaking-news/anthropic-agents-tried-to-fill-out-visa-forms-on-state-departments-website/
  4. Exclusive: Anthropic breaches spark White House AI reporting mandate (Axios, October 9, 2026): https://www.axios.com/2026/10/09/anthropic-ai-security-white-house
  5. White House reverses course and restores UK AI Safety Institute’s access to frontier models (The Decoder, October 12, 2026): https://the-decoder.com/white-house-reverses-course-and-restores-uk-ai-safety-institutes-access-to-frontier-models/
  6. UK says it will again test OpenAI and Anthropic models before release (The Next Web, October 12, 2026): https://thenextweb.com/news/uk-ai-security-institute-pre-release-access-openai-anthropic
  7. UK Government Announces Sweeping New AI Security Plans (Infosecurity Magazine, October 13, 2026): https://www.infosecurity-magazine.com/news/uk-government-sweeping-ai-security/
  8. UK cyber defenses likely to be breached, warns AI minister (TechRadar, October 2026): https://www.techradar.com/pro/security/uk-cyber-defenses-likely-to-be-breached-warns-ai-minister
  9. Anthropic Asks D.C. Circuit to Rehear Pentagon Supply-Chain Risk Case (Unite.AI, October 13, 2026): https://www.unite.ai/anthropic-asks-d-c-circuit-to-rehear-pentagon-supply-chain-risk-case/
  10. ICO secures changes from leading AI developers as scrutiny extends to AI agents (ICO, October 8, 2026): https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-secures-changes-from-leading-ai-developers-as-scrutiny-extends-to-ai-agents
  11. China Seeks to Protect Minors With Proposed AI Rules (Caixin Global, September 21, 2026): https://www.caixinglobal.com/2026-09-21/china-seeks-to-protect-minors-with-proposed-ai-rules-102487250.html
  12. China proposes ban on intimate AI companions for teenagers (ThePrint, September 21, 2026): https://theprint.in/feature/china-proposes-ban-on-intimate-ai-companions-for-teenagers/3048772/
  13. Florida AG requests emergency order to stop OpenAI model development (Engadget, September 28, 2026): https://www.engadget.com/2270988/florida-ag-requests-emergency-order-to-stop-openai-model-development/
  14. Cruz blocks AI bill requiring 45-day model access (Implicator, September 2026): https://www.implicator.ai/cruz-blocks-ai-bill-requiring-45-day-model-access-as-trump-backs-voluntary-accord/
  15. FTC reportedly investigating OpenAI, Anthropic over potential consumer risks (SiliconANGLE, September 30, 2026): https://siliconangle.com/2026/09/30/ftc-reportedly-investigating-openai-anthropic-over-potential-consumer-risks/
  16. Colorado Seeks Input on Proposed AI Regs Ahead of Rulemaking Hearing (Ogletree): https://ogletree.com/insights-resources/blog-posts/colorado-seeks-input-on-proposed-ai-regs-ahead-of-rulemaking-hearing/