Anthropic Accidentally Leaked Its Secret Superweapon and Now Everyone Knows About 'Mythos'

Oops. Anthropic, the company behind the popular Claude AI, just had one of the most embarrassing leaks in tech history. A new AI model called "Mythos" was accidentally left sitting in a publicly accessible data store where anyone could find it.

Here's what happened: someone at Anthropic misconfigured their content management system (basically, they left the door unlocked), and a draft blog post announcing Claude Mythos ended up visible to the entire internet. Security researchers found it, Fortune reported on it, and now the cat is very much out of the bag.

What makes this a big deal? Anthropic themselves called Mythos a "step change" in performance and "the most capable model we've built to date." The leaked draft also revealed something alarming: the company believes this model poses "unprecedented cybersecurity risks." That's the company that built it saying that.

The leak also exposed plans for a secret, invite-only CEO summit in Europe where Anthropic planned to pitch Mythos to Fortune 500 companies. Nearly 3,000 unpublished documents were sitting in that unsecured data cache before Anthropic locked it down.

Anthropic blamed "human error" and says the model is currently being tested with early access customers. But the irony of an AI safety company having a major security lapse is not lost on anyone.

As reported by Fortune.

Source: Fortune