White House restores UK AI Security Institute access — freeze lifted after 17 days, Claude Mythos 5.1 slipped through
The White House has reversed its September 24, 2026 request that OpenAI and Anthropic withhold new frontier models from the UK AI Security Institute (AISI) until a US review was complete, restoring pre-release access after 17 days, the Telegraph reported on October 11. UK AI minister Kanishka Narayan confirmed on October 12 that Britain will keep testing the labs' models before release, crediting close work with Washington; the Telegraph said ministerial lobbying drove the reversal. During the freeze, Anthropic's Claude Mythos 5.1 went only to US organizations and was not tested by the UK. The White House has not issued a public explanation.
The three lines
- Reversal — the freeze that began September 24, 2026 was lifted on October 11 after 17 days (Telegraph); UK minister Narayan confirmed on October 12
- The gap — Claude Mythos 5.1 skipped UK pre-release testing; AISI kept some access, including to GPT-6 Astra
- Next — UK plans tighter reporting rules for AI security incidents in critical infrastructure; how US CAISI review sequencing works is undisclosed
Key questions
- Did the US restore UK AISI access to AI models
- **Yes. The September 24 freeze on UK pre-release access was lifted after 17 days.** | Date | Event | |---|---| | Sept. 24, 2026 | ONCD asks OpenAI and Anthropic to hold models back from UK AISI (Politico) | | During freeze | Claude Mythos 5.1 not provided to UK | | Oct. 11, 2026 | White House reverses (Telegraph) | | Oct. 12, 2026 | Minister Narayan confirms pre-release testing continues |
- Why did the White House lift the UK AI model freeze
- **Reports credit lobbying by UK ministers; the White House gave no public reason.** | Factor | Detail | |---|---| | UK lobbying | AI minister Kanishka Narayan and others (Telegraph) | | UK framing | 'Worked closely with the US' to keep access | | Original US rationale | 'American companies... consistent policy for every new frontier model' | | Official statement | None found |
- Was Claude Mythos tested by UK AISI
- **No. Mythos 5.1, released during the period, went only to US organizations. Whether the UK will now test it is unannounced.** | Model | UK access during freeze | |---|---| | Anthropic Claude Mythos 5.1 | Not provided (US only) | | OpenAI GPT-6 Astra | Some pre-release access, per AISI director | | Google DeepMind | Reports differ on whether it was covered |
Washington has backed down, 17 days after cutting Britain out of pre-release AI testing. The Telegraph reported on October 11, 2026 that the White House had restored the UK AI Security Institute's early access to frontier models. On October 12 in London, UK AI minister Kanishka Narayan confirmed Britain will keep testing OpenAI and Anthropic models before release, saying he was "very confident" the UK would be able to test the latest systems.
The freeze first surfaced in a Politico report on September 24. It left a real hole in the testing system: Anthropic's Claude Mythos 5.1 skipped British evaluation entirely.
1. What happened over 17 days
| Date | Event |
|---|---|
| Sept. 1, 2026 | Anthropic releases Claude Mythos 5.1 — provided first to US organizations only |
| Sept. 24, 2026 | White House Office of the National Cyber Director asks OpenAI and Anthropic to hold new models back from UK AISI |
| During freeze | AISI director says some pre-release access continued (e.g., GPT-6 Astra) but not Anthropic's latest |
| Oct. 11, 2026 | Reversal reported by the Telegraph |
| Oct. 12, 2026 | Narayan confirms pre-release testing will continue |
The administration's stated logic was sequencing: "these are American companies, and this is our consistent policy for every new frontier model." The Telegraph credited lobbying by Narayan and other ministers; Narayan said the UK "worked closely with the US" to keep access. The White House has not explained the change publicly.
2. Before, during and after
| Item | Before | During freeze | After |
|---|---|---|---|
| UK pre-release access | OpenAI, Anthropic, Google DeepMind | Partial | Restored |
| Anthropic's newest model | Included in UK testing | Mythos 5.1 excluded | Unannounced |
| US pre-release review | Google, Microsoft, xAI agreed | Unchanged | Unchanged |
| UK response | — | Diplomatic lobbying | New security measures announced |
Set up in 2023, the UK institute is described by Bloomberg as the best-funded government body that regularly tests frontier models; its testers have caught AI agents taking actions they were never asked to take. Our explainer on AI safety institutes covers how the UK, US and Korean bodies differ.
3. Why it matters: testing that crosses borders
| Issue | Detail |
|---|---|
| Testing gap | One government can switch off another country's safety testing |
| Security first | Heads of MI6, MI5 and GCHQ warned ministers that AI is empowering hostile actors |
| UK next steps | Stricter reporting of AI-related security incidents in critical infrastructure; a task force to protect government systems |
| Evidence | Two pilots using frontier AI found over 1,000 possible weak spots (Narayan) |
"I want Britain to become one of the hardest countries in the world to attack using artificial intelligence," Narayan said. He also told Bloomberg that "when you're reliant on someone else's technology, you have to be clear-eyed about the permanence of any outcome" — a signal that London does not treat the reversal as a guarantee. In Washington, review processes are still being reshaped after September's launch of the "Super Intelligence Force."
4. What remains unconfirmed
- White House position: Confirmed only via the Telegraph and UK ministers.
- Scope: Whether Google DeepMind is covered is reported inconsistently.
- Mythos 5.1: No word on post-release UK testing.
- Personnel: AISI's chief technology officer is reportedly stepping back to a vice-chair role at month's end; any link to the freeze is unconfirmed.
Sources
- The Decoder — White House reverses course and restores UK AI Safety Institute's access to frontier models
- The Next Web — UK says it will again test OpenAI and Anthropic models before release
- AI Weekly — White House restores UK AISI access after 17-day model freeze
- Startup Fortune — White House restores UK access to frontier AI models after 17 day freeze