US government re-authorises Anthropic’s AI models
The USA Department of Commerce has lifted export controls on Mythos 5 and Fable 5. The two AI models from Anthropic were banned three weeks ago because they were deemed too dangerous.
Following a three-week enforced hiatus, Anthropic is once again permitted to offer its AI models, Fable 5 and Mythos 5. The US Department of Commerce has lifted the export controls that had been imposed in mid-June. Anthropic announced that it would gradually restore access to Fable 5 worldwide on its own Claude platforms from Wednesday. Mythos 5 will remain restricted to selected partners for the time being.
Previously, the government had required the company to block access to both models for foreign nationals – including those within the USA. As Anthropic was unable to implement this requirement reliably from a technical standpoint, the company took Fable 5 and Mythos 5 offline globally. The authorities justified the move on the grounds of security concerns: researchers, including those at Amazon, had demonstrated that, despite its protective mechanisms, Fable 5 could be used as a tool for cybercrime.
Open model from China increases the pressure
In a blog post, Anthropic describes the conditions under which the models are being reinstated. An improved security classifier is designed to block the known jailbreak method in over 99 per cent of cases. If a request is intercepted, the user is informed and the prompt is forwarded to the weaker model, Opus 4.8.
At the same time, Anthropic is expanding its cooperation with the US government. Government agencies are to be granted advance access to security-relevant models. The company intends to inform the government more quickly in the event of newly discovered jailbreaks. According to the blog post, Anthropic is also working with partners such as Amazon, Microsoft and Google on a framework to assess jailbreaks in a standardised manner.
At the same time, international competitive pressure is mounting. The Chinese company Zhipu AI has released an open-weight model called GLM‑5.2. Although it is said to lag behind models from Anthropic and OpenAI on general tasks, it is claimed to be able to compete with Mythos in the field of cybersecurity.
The release at least partially defuses the conflict between Anthropic and Washington. Anthropic regains a prominent AI model in the run-up to its planned IPO. At the same time, the government reserves the right to roll it out in stages, and Mythos 5 will initially be restricted to a pre-approved list of organisations. For companies in Europe, this highlights just how dependent they are on regulatory decisions in the USA.
My fingerprint often changes so drastically that my MacBook doesn't recognise it anymore. The reason? If I'm not clinging to a monitor or camera, I'm probably clinging to a rockface by the tips of my fingers.
From the latest iPhone to the return of 80s fashion. The editorial team will help you make sense of it all.
Show all
