Anthropic investigates claim of unauthorised access to Mythos AI tool

The article notes that frontier AI models are developed outside the UK, with top-tier companies based in the US or China, and that the UK relies on companies like Anthropic to gain access to Mythos. It also mentions OpenAI’s cyber-security model GPT 5.4 Cyber and ongoing threats from nation-states and hacktivists.

Security Minister Dan Jarvis called on AI firms to work with the government in a generational effort to ensure AI is used to protect critical networks from attackers.

At CyberUK, Richard Horne urged delegates to focus on basics of cyber-security and argued frontier AI can be safer and more secure if underlying practices are strengthened.

Bloomberg reported that the person who accessed Mythos already had permission to view Anthropic’s AI models through work for a third-party contractor, and that the group had been using the model since gaining access.

Anthropic has released the Mythos model to some tech and financial companies to assist in securing their systems against its reported ability to exploit vulnerabilities, with the caveat that those companies must tightly control access.

Raluca Saceanu, chief executive of Smarttech247, described the incident as most likely arising from misuse of access rather than a traditional hack.

The company said there is no evidence that its systems were compromised and there is no indication that malicious actors currently hold the model.

Anthropic issued a statement saying it is investigating a report that unauthorised access to Claude Mythos Preview occurred through a third-party vendor environment.

Bloomberg reported that a small group of people in a private forum were able to access Claude Mythos Preview without the normal permissions. The report notes the group had been using the model since gaining access and tried to avoid detection, with claims that the individual had permission to view Anthropic’s AI models through a third-party contractor.