Anthropic Opens Its Most Powerful AI Model to the Public, Strengthens Safety Measures

Anthropic, the developer of the artificial intelligence (AI) model Claude, on Tuesday opened its most powerful version of the technology to the public while restricting its use in sensitive areas.

The model, called 'Fable 5,' is the first widely available model in Anthropic's new 'Mythos' series. Mythos is Anthropic's most advanced AI technology product line, announced in April, but access was limited due to cybersecurity concerns.

Anthropic said Fable 5 excels at writing and debugging software code, answering complex research questions, and analyzing images.

At the same time, Anthropic is providing an unrestricted version, 'Claude Mythos 5,' to companies and organizations that already have access to the model series. This includes cybersecurity partners participating in 'Project Glasswing.'

According to Reuters, after Anthropic announced in April that Mythos had discovered thousands of software vulnerabilities, access was limited to about 200 organizations, including the U.S. government, under 'Project Swallowtail.'

This select group was expanded in early June to about 200 organizations across more than 15 countries, and is expected to grow further.

Anthropic has restricted access to the model due to security concerns, citing Mythos's ability to quickly identify vulnerabilities in critical infrastructure such as banking platforms and power grids.

When 'Project Glasswing' was launched, some critics accused Anthropic of exaggerating the threat to attract attention.

However, companies that tested Mythos have since acknowledged the model's capabilities. The U.S. government, which had a legal dispute with Anthropic, also tested the model due to security concerns.

The White House has since arranged for testing of the most powerful models before they are released by leading AI companies. Mythos 5 is being deployed in collaboration with the U.S. government.

Anthropic stated that most queries related to security or biology and chemistry will be redirected to a less capable model, 'Opus 4.8,' which was released at the end of May.

Anthropic also said it has detected attempts to massively extract related technology to train competing AI models in authoritarian countries, and such queries will similarly be redirected to the less capable model.