Anthropic releases Claude Mythos cybersecurity model to select firms

In short: Anthropic has started a limited release of Claude Mythos, an AI model designed to find hidden security problems in software, to a small group including Amazon, Apple, and Microsoft.

What happened

Anthropic said it is rolling out “Claude Mythos Preview” to vetted organisations only. The initial group includes Amazon, Apple, Microsoft, and security companies such as Broadcom, Cisco, and CrowdStrike.

The release comes shortly after two internal security incidents at Anthropic. Last month, descriptions of the Mythos project and other documents were found in a publicly accessible online storage area. Then a second incident made internal source code for “Claude Code” public. Source code is the human written instructions behind software (like a recipe that tells a computer what to do). Anthropic said both incidents were caused by human error.

Anthropic said Mythos has been used with partners for several weeks. The company describes it as a general purpose AI model, but with unusually strong skills in cybersecurity. Cybersecurity means protecting computers and networks from break ins, like locks and alarm systems for digital systems.

Why it matters

Anthropic is limiting access because the same tool that can find weak spots could also help someone abuse them. Mythos has reportedly found thousands of “zero-day” vulnerabilities, which are security holes that were not known before (like finding a hidden door in a building). In one example, it found a 16 year old flaw in widely used video software.

Anthropic also said it is talking with the US government about potential use. The company plans to offer up to $100 million in credits to participating organisations and donate $4 million to open source security groups.

Source: Financial Times