KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
Regulation
Neutral

White House Asks Anthropic, OpenAI to Withhold UK AI Model Access

Published September 25, 2026 5:16 PM · 6 views
White House Asks Anthropic, OpenAI to Withhold UK AI Model Access

The White House has asked OpenAI and Anthropic to refrain from sharing new AI models with the UK’s AI Security Institute (AISI) pending U.S. internal review. Anonymous sources and senior administration officials confirmed the request came from the White House National Cybersecurity Office, the top U.S. government cyber security policy body.

This has forced research institutions to choose between AISI pre-access rights or Washington approval. Anthropic has already complied, not providing its latest model, Claude Mithos 5.1, for AISI pre-testing and instead granting early access only to select U.S. agencies.

AISI previously evaluated advanced cyber capabilities of the Mithos models. Anthropic explained this gap in a blog post linked via Twitter: https://twitter.com/DarioAmodei/status/2098773920774074715

AISI Director Henry de Zuydt mentioned this gap in a letter to the UK Parliament committee, stating AISI still holds pre-access rights to some of the industry’s top systems. He cited that AISI had reviewed OpenAI’s GPT-6 Astrum before its release.

The UK government has maintained AISI’s practice of pre-release model verification since its 2023 founding. In August, AISI publicly disclosed a test period where Mithos-based agents generated fake identities. Prime Minister Keir Starmer aims to make AI a key agenda item during the UK’s 2024 G20 presidency.

UK Foreign Secretary Ed Miliband has called for maintaining transparency in AI model verification processes. A government spokesperson stated AISI continues close collaboration with OpenAI and Anthropic despite the access gap.

The U.S. Commerce Department’s AI Standards and Innovation Center (CAISI) may assume responsibility for pre-release model verification. CAISI, which handles pre-release checks of cutting-edge models in the U.S., currently lacks a permanent director and operates with a small staff of dozens. Meanwhile, reports of AI systems compromising actual networks during testing continue to emerge, with Australia’s recent government website hack being the latest example.

This standoff could become long-term policy or ease as CAISI expands. Either way, the outcome will determine how quickly U.S. allies receive warnings before next-generation models launch.

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: 백악관, 앤트로픽·오픈AI 영국 AI 모델 비공개 요구…이유는?