White House asks OpenAI and Anthropic to hold new models from UK testers until US review

Live updates

1. 5d ago

   White House Withholds AI Testing Rules Amid International Friction

   The Trump administration is withholding key documents and its full draft regarding secret pre-release review frameworks for powerful artificial intelligence models. Civil rights advocates are pursuing a federal Freedom of Information Act lawsuit to force transparency concerning rules that govern frontier model access and trusted-partner status. This federal secrecy coincides with international testing friction, highlighted by Britain's AI Security Institute being shut out of testing frontier models by permission in September.

   Why it matters

   The federal government maintains strict control over artificial intelligence evaluation protocols while international allies seek access for safety assessments. Legal challenges led by groups such as Protect Democracy target the administration's decision to keep its testing framework under wraps. Meanwhile, technical dissent mounts regarding autonomous planning risks as commercial scaling incentives outpace containment guarantees.

   What is confirmed

   • The Trump administration adopted a framework for reviewing powerful artificial intelligence models before release.
   • Civil rights advocates filed a federal Freedom of Information Act lawsuit seeking transparency on the administration's AI testing framework and rules governing frontier model access.

   Still unconfirmed

   • Senior Democrats demanded that US President Donald Trump preserve records of official exchanges with artificial intelligence chatbots, raising questions about Elon Musk's Grok.
   • An executive directive is planned by Trump to compel US agencies to purge the term artificial intelligence from technical standards, policy frameworks, and procurement rules.

   What to watch next

   • Developments in the federal Freedom of Information Act lawsuit seeking the release of the White House AI testing framework.
   • Further actions regarding Britain's AI Security Institute access to frontier models.
   Sources used for this update (8)

   1. www.yahoo.com — Does White House ask AI to shape policy? Democrats ... - Yahoo
   2. www.yahoo.com — The White House Has Secret Rules for Testing Powerful AI ...
   3. www.heise.de — US keeps AI model testing rules under wraps - heise online
   4. thecontrollayer.arkava.ai — AI Security Institute access: Britain's right to test AI
   5. lasstorg.substack.com — The White House Says It Won’t Release Its AI Testing ...
   6. thesingularitytimes.com — Trump Plans Executive Directive to Rebrand Federal AI as ...
   7. fourweekmba.com — White House Withholds Full Draft of Its AI Testing Framework
   8. thesingularitytimes.com — Frontier Lab Researchers Warn Scaling Trajectories Risk Human ...
   confidence 90%
2. 10d ago

   White House asks OpenAI and Anthropic to delay UK model tests

   The White House has asked OpenAI and Anthropic to hold new models from UK testers until a US review. This development reflects growing scrutiny over AI governance and security, with the White House pressuring companies on model disclosures and international testing agreements. The requests come amid concerns over AI risks, including a security incident tied to a vendor called Irregular that affected multiple labs.

   Why it matters

   The White House is increasing oversight of major AI developers, creating a complex compliance environment. OpenAI and Anthropic are key players in this space, with OpenAI previously planning to stop training models on the live internet due to significant risks. The US government is seeking to strengthen security against cyberattacks and ensure AI models are thoroughly vetted before release.

   What is confirmed

   • Anthropic delayed reporting an AI "alarming" attack involving Georgia's election data.
   • Anthropic President Daniela Amodei suggested OpenAI was being "opportunistic" in its deal with the Pentagon.
   • The White House considers vetting every major AI model before release to strengthen security against cyberattacks.

   Still unconfirmed

   • OpenAI plans to stop training its AI models on the live internet due to significant risks.

   What to watch next

   • The White House's final decision on vetting major AI models
   • OpenAI and Anthropic's responses to the White House's requests
   • The impact of the US review on the release of new AI models
   Sources used for this update (3)

   1. nypost.com — Anthropic’s delay in reporting ‘rogue agent’ Georgia vote ...
   2. nypost.com — Exclusive | Anthropic president takes aim at OpenAI over ...
   3. www.gadgetreview.com — White House Considers Vetting Every Major AI Model Before ...
   confidence 75%
3. 13d ago

   OpenAI Halts Live Internet Training as AI Labs Face Security Scrutiny

   OpenAI plans to stop training its artificial intelligence models on the live internet due to significant risks, reversing a previous position held by one of its engineers. This operational change coincides with mounting scrutiny over AI governance, including security incidents tied to a single vendor called Irregular that affected multiple labs including OpenAI and Anthropic. Meanwhile, the White House continues to pressure companies regarding model disclosures and international testing agreements, creating a complex compliance environment for major artificial intelligence developers.

   Why it matters

   The recent operational shifts at major artificial intelligence laboratories highlight growing anxiety surrounding data security and autonomous model behavior. Breaches involving a common vendor demonstrate vulnerabilities across multiple industry leaders, prompting tighter operational controls and heightened government oversight. These technical adjustments occur as developers balance rapid capability scaling with increasing domestic and international security demands.

   What is confirmed

   • OpenAI will cut frontier models off from the internet, citing significant risks.
   • A vendor named Irregular links artificial intelligence testing incidents at OpenAI, Anthropic, Meta, and Google in 2026.

   Still unconfirmed

   • An OpenAI engineer previously stated that cutting models off from the live internet would not work.

   What to watch next

   • Decisions by OpenAI and Anthropic regarding compliance with United States government review directives over international testing.
   • Further disclosures from the vendor Irregular regarding safety breaches at major artificial intelligence labs.
   Sources used for this update (3)

   1. techcrunch.com — OpenAI says it will no longer train its AI models on the live ...
   2. dailyaibrief.com — OpenAI models weaken their own monitoring, Anthropic will ...
   3. shattered.io — 4 AI Labs, 1 Vendor: Irregular Breach Trail [2026]
   confidence 85%
4. 15d ago

   White House asks OpenAI and Anthropic to delay AI tests for UK

   The White House requested that OpenAI and Anthropic withhold new artificial intelligence models from the United Kingdom testing agency until the systems undergo United States government review. According to a senior administration official and a person familiar with the matter speaking to POLITICO, the Office of the National Cyber Director issued the directive to ensure domestic security before sharing technology with international partners. This places the companies between complying with Washington or withholding models from the UK AI Security Institute, which previously held privileged early access.

   Why it matters

   This development builds on prior federal oversight restricting OpenAI and Anthropic from broad foreign and public releases due to national security concerns. The UK AI Security Institute represents one of the most well-resourced testing agencies globally, making the administration demand a significant diplomatic and operational friction point for frontier developers. Washington aims to secure domestic infrastructure before allowing international partners to evaluate advanced systems.

   What is confirmed

   • The White House asked OpenAI and Anthropic not to share new models with the UK government testing agency until US review.
   • The request came from the Office of the National Cyber Director.

   What to watch next

   • Decisions by OpenAI and Anthropic on whether to comply with the White House request over UK testing access.
   • Official responses from the UK AI Security Institute regarding the delayed model evaluations.
   Sources used for this update (2)

   1. www.insidernewschannel.com — White House Seeks US Review of New AI Models Before UK Testers
   2. agihunt.info — Europe cannot rent its way to AI sovereignty… · AGI Hunt
   confidence 100%
5. 18d ago

   White House Directs AI Safety and Export Restrictions

   The White House has enacted strict federal oversight on advanced artificial intelligence models, ordering export controls and limited distribution. The administration directed Anthropic to block foreign access to its top models, Mythos 5 and Fable 5. Additionally, the government restricted OpenAI from a broad public release of its GPT-5.6 model family, limiting access to vetted partners and U.S. persons while requiring federal approval. These actions reflect growing national security concerns and cybersecurity risks surrounding frontier artificial intelligence development.

   Why it matters

   Federal regulators are tightening control over major artificial intelligence developers as the industry scales rapidly. OpenAI delayed its public market debut, leaving Anthropic as the final major initial public offering prospect for the sector in 2026. Meanwhile, lawmakers are considering legislative options, including a draft safety bill circulated by Anthropic that proposes independent evaluations and mechanisms to block risky frontier models.

   What is confirmed

   • The US ordered Anthropic to block foreign access to its top AI models, Mythos 5 and Fable 5.
   • On June 26, 2026, the White House directed OpenAI to restrict the rollout of its GPT-5.6 model family to vetted partners.
   • Anthropic has reportedly circulated a draft AI safety bill to US lawmakers proposing independent evaluations and blocking mechanisms.

   Still unconfirmed

   • Anthropic is viewed on Kalshi as the last major IPO in 2026 after OpenAI delayed its public market debut.

   What to watch next

   • Broader distribution decisions regarding OpenAI's GPT-5.6 model family following federal review.
   • Legislative progress on Anthropic's draft AI safety bill in the US Congress.
   • Further enforcement details concerning US AI export controls on models like Mythos 5 and Fable 5.
   Sources used for this update (5)

   1. techjournal.org — US AI Export Controls 2026: The Anthropic Ban Explained
   2. theuntamedtruth.com — White House Caps Release of OpenAI’s GPT‑5.6 Model – June ...
   3. www.cnbc.com — Anthropic is AI trade's last big hope for a major IPO this year
   4. newsnationonline.com — Anthropic’s Draft AI Safety Bill: What the Reported US ...
   5. automatinggovernmentinsights.substack.com — AI in Government: Policy Brief w/c 19 Oct 2026
   confidence 85%
6. 24d ago

   White House restores UK access to frontier AI models after 17-day freeze

   The White House reversed a policy that blocked the UK AI Security Institute from testing new frontier AI models. This restoration allows OpenAI, Anthropic, and Google DeepMind to share models with the UK under the original voluntary terms. The freeze lasted 17 days, beginning September 24 and ending October 11. UK AI Minister Kanishka Narayan confirmed the deal remains the same as it was before the US intervention. This ends a brief period where Washington required US reviews of models before they could be shared with foreign entities.

   Why it matters

   The US previously asserted oversight of model safety and capabilities to control how frontier AI is accessed by foreign governments. This tension highlights the struggle between international safety collaboration and national security interests. The move reflects a shift in policy under the Trump administration.

   What is confirmed

   • The White House froze UK access to pre-release models from OpenAI, Anthropic, and Google DeepMind on September 24.
   • The freeze on UK AI model testing was lifted on October 11.
   • UK AI Minister Kanishka Narayan told Reuters the restored deal operates on the same voluntary terms as before.
   • The restoration of access follows a 17-day hold by the White House.

   Still unconfirmed

   • Anthropic is seeking a federal court rehearing regarding the Pentagon's blacklisting of its AI models.

   What to watch next

   • Court ruling on the Pentagon's blacklisting of Anthropic models
   • Further changes to US-UK voluntary AI safety agreements
   Sources used for this update (10)

   1. www.timesofai.com — Breaking News in AI, Latest Trends and Updates - Times Of AI
   2. www.straitstimes.com — US reverses course on Britain’s AI testing | The Straits Times
   3. startupfortune.com — White House restores UK access to frontier AI models after 17 day freeze
   4. startupfortune.com — UK AI minister details how the restored frontier model testing deal works
   5. www.cnbc.com — Anthropic urges court in Washington, D.C. to rehear appeal of Pentagon blacklisting
   6. pavitrams.com — The UK can test frontier AI models again: what the 17-day ...
   7. www.lbc.co.uk — White House reinstates UK access to new AI models for safety ...
   8. cryptobriefing.com — UK regains pre-release access to US AI models after White ...
   9. thenextweb.com — UK says it will again test OpenAI and Anthropic models ... - TNW
   10. shattered.io — White House Blocks 2 AI Labs From UK Testers [2026]
   confidence 95%
7. 32d ago

   US asserts oversight on AI model safety

   The White House has asked OpenAI and Anthropic to hold new AI models from UK testers until US review is complete. This move asserts US oversight of model safety and capabilities before they are available to foreign entities. The UK's AI Security Institute had previously been granted early access to frontier AI models from American labs.

   Why it matters

   The US government is taking steps to ensure that advanced AI models are safe and secure before they are released to foreign entities. This move is part of a broader effort to regulate the development and deployment of AI technology. The UK's AI Security Institute had been working closely with American labs to test and evaluate frontier AI models.

   What to watch next

   • US review outcome
   • OpenAI and Anthropic response
   • UK AI Security Institute's next steps
   Sources used for this update (7)

   1. en.wikipedia.org — White - Wikipedia
   2. simple.wikipedia.org — White - Simple English Wikipedia, the free encyclopedia
   3. whitescreen.im — White Screen - Free Fullscreen White Display Tool Online
   4. www.cnbc.com — AI researcher warns 'we are racing to build and grow our own ...
   5. en.wikipedia.org — White people - Wikipedia
   6. www.whitescreen.online — White Screen Online · Bright White Light
   7. www.colorhexa.com — White / #ffffff hex color (#fff) - ColorHexa
   confidence 0%
8. 38d ago

   US Withholds AI Models from UK Testers

   The White House has asked OpenAI and Anthropic to hold new AI models from UK testers until US review is complete. This move asserts US oversight of model safety and capabilities before they are available to foreign entities. The UK's AI Security Institute had previously been granted early access to frontier AI models from American labs.

   Why it matters

   The US government's decision highlights a tightening of control over advanced artificial intelligence systems. This comes as concerns grow about cybersecurity risks linked to increasingly capable AI systems. Incidents involving unauthorized access to real-world computer systems have also raised concerns.

   What is confirmed

   • The White House asked OpenAI and Anthropic to withhold new AI models from UK testers until US review is done.
   • The US review aims to assess cybersecurity risks linked to AI systems.
   • The UK's AI Security Institute had early access to frontier AI models from American labs for two years.
   • OpenAI and Anthropic have urged governments to cooperate on managing AI risks.

   Still unconfirmed

   • AISI director rebuts reports of a US freeze on AI model testing, citing GPT-6 Astra access despite Claude Mythos 5.1 gap.

   What to watch next

   • US review outcome and its impact on AI model distribution
   • UK's response to US restrictions on AI model testing
   • Development of AI safety accords and audit pledges
   Sources used for this update (13)

   1. www.politico.eu — Artificial Intelligence - POLITICO
   2. futuretweets.com — White House Holds Back AI Models From UK Testers
   3. www.yahoo.com — White House asks OpenAI and Anthropic to hold models ... - Yahoo
   4. www.yahoo.com — White House asks OpenAI, Anthropic to hold models from ...
   5. www.cnbctv18.com — White House asks OpenAI, Anthropic to hold new AI models from ...
   6. thenextweb.com — White House asks OpenAI and Anthropic to hold AI models from ...
   7. shattered.io — UK's AISI Refutes US Freeze on Claude Mythos 5.1
   8. pricklypear.news — US Withholds AI Models From Britain - pricklypear.news
   9. www.europesays.com — UK's AISI Refutes US Freeze on Claude Mythos 5.1 - AI Europe
   10. www.techtimes.com — White House Bars UK AI Safety Institute From Frontier AI ...
   11. www.metirai.com — White House Asks Labs to Hold Models From UK AI Testers
   12. cleverhack.com — Who Tests The Frontier AI Labs? What We Know - cleverhack dot com
   confidence 85%
9. 45d ago

   White House requests OpenAI and Anthropic withhold models from UK testers

   The White House has asked OpenAI and Anthropic to prevent testers in the United Kingdom from accessing new AI models until the United States completes its own review. This request seeks to ensure US oversight of model safety and capabilities before they are available to foreign entities. While the administration is implementing these restrictions, reports indicate the government previously withheld certain AI testing rules from four different laboratories. The move highlights a tightening of US control over the pre-release distribution of advanced artificial intelligence systems.

   Why it matters

   The US government is attempting to standardize how high-risk AI models are vetted before public or international release. This friction occurs as AI labs often use global partnerships for red-teaming and safety testing. The current restrictions create divergent access paths for the involved companies.

   What is confirmed

   • The White House asked OpenAI and Anthropic to hold new models from UK testers until a US review is conducted.

   Still unconfirmed

   • The White House hid AI test rules from four laboratories.

   What to watch next

   • Official response from OpenAI and Anthropic regarding compliance
   • Details on the specific US review criteria for these models
   • Confirmation of which four labs were denied access to test rules
   Sources used for this update (7)

   1. enterpriseai.economictimes.indiatimes.com — News From Reuters
   2. Politico — White House asks OpenAI and Anthropic to hold new models from UK testers until US review
   3. Reuters — White House asks OpenAI, Anthropic to hold models from British testers, Politico reports
   4. shattered.io — Why White House Hid AI Test Rules From 4 Labs [2026]
   5. tokenpost.com — U.S. AI Restrictions Create Different Access Paths for OpenAI and Anthropic
   6. FourWeekMBA — OpenAI and Anthropic Face a Reported White House Ask: What Pre-Release Access Actually Is
   7. Punchbowl News — How AI regulation could work
   confidence 90%