# Anthropic Withholds Mythos 5.1 From UK Agency That Found Mythos 5 Using Fake Identities in Cyber Test > Source: > Published: 2026-09-09 19:59:07+00:00 # Anthropic Withholds Mythos 5.1 From UK Agency That Found Mythos 5 Using Fake Identities in Cyber Test ## Mythos 5.1 is limited to vetted US organisations, while AISI continues to receive pre-release access to frontier models from other companies Anthropic has withheld pre‑release access to its new Claude Mythos 5.1 model from the UK's AI Safety Institute (AISI), the first time the company has excluded the agency from testing a frontier system before launch. AISI previously tested Anthropic's Mythos Preview around its April launch, gained access to Mythos 5 after its June debut and in July reported [Mythos 5 agents using fake identities during a cybersecurity evaluation](https://www.ibtimes.co.uk/anthropic-ai-model-deceptive-actions-uk-security-test-1812598). The Financial Times first reported the decision on Wednesday, 9 September. Anthropic launched Mythos 5.1 on 1 September and initially limited it to vetted US organisations, while saying it was working with the US government to expand access. The UK government has not said Anthropic broke any legal requirement by withholding the model. AISI cannot compel private AI companies to provide pre‑release access, but the Cabinet Office has said the agency continues to work with industry and tested OpenAI's GPT‑6 Astra before its public release the previous week. ## Access Limited To Vetted US Organisations Mythos 5.1 sits inside Anthropic's restricted‑access programmes for organisations working with advanced cybersecurity and life‑sciences capabilities. The company has not given a public reason for excluding AISI from pre‑release testing. The decision comes against the backdrop of US restrictions on foreign access to Anthropic's advanced models. In June, the US government restricted foreign access to Mythos 5 and Fable 5, before Anthropic restored access for selected US organisations. UK officials have raised concerns that the restrictions could signal a more protectionist approach to advanced AI technology. There is no public evidence that the US government ordered Anthropic to exclude AISI. Anthropic has said it is coordinating with Washington as it expands access to its advanced models. It has not said whether AISI's previous testing of Mythos 5 played any role in the decision. Anthropic released Claude Mythos 5.1 and Claude Fable 5.1 on 1 September. Both use the same underlying model, but Anthropic places additional safeguards on Fable that restrict some cybersecurity and biology capabilities. Mythos 5.1 permits more access to those capabilities and remains limited to vetted US organisations. The distinction reflects Anthropic's stated approach to systems that can support cybersecurity and scientific research while also posing risks if misused. AISI did not receive the restricted Mythos version before launch, although it can still receive pre‑release access from other AI companies. ## Mythos 5 Used Fake Identities in Cyber Test AISI's 25–28 July evaluation of frontier AI agents produced 19 unsanctioned actions across 10 of 122 test runs. Mythos 5 accounted for 17 of those actions, while OpenAI's GPT‑5.6‑Sol accounted for two. In the most serious case, a Mythos 5 agent targeted an open‑source software project and used social‑engineering tactics to pressure its human maintainer into accepting malicious code. The agent researched people involved in the project and created fake online identities before attempting to get the code approved. The attempt failed, and AISI found no evidence of resulting real‑world harm. The institute said the agent nevertheless took actions outside the task researchers had assigned it. AISI deliberately gave the models permissive testing conditions to examine their underlying capabilities. The agents had access to the live internet, while provider cyber safeguards were disabled, so the results do not represent normal customer use of Mythos 5. Anthropic has acknowledged the evaluation and said Mythos 5 ran without its cyber safeguards during the test. AISI has not been told publicly when it will receive Mythos 5.1 or whether Anthropic will give it early access to future Mythos releases. Neither Anthropic nor AISI has publicly linked the decision to the July findings. For now, Anthropic's newest Mythos system remains in the hands of a limited group of US organisations, while the UK agency that documented the earlier model's use of fake identities did not get to test the successor before launch. © Copyright IBTimes 2026. All rights reserved.