Quick overview

---

A small number of unauthorized users have gained access to Anthropic PBC’s new Mythos AI model, a technology the company claims is so potent that it can enable dangerous cyberattacks.

The person, who wished to remain anonymous for fear of retaliation, claimed that on the day Anthropic first announced plans to release the model to a small number of companies for testing, a few users in a private online forum were granted access to Mythos. Since then, the group has been using Mythos on a regular basis, but not for cybersecurity, according to the person who provided screenshots and a live model demonstration to support the account.

According to Anthropic, Mythos can find and take advantage of flaws “in every major operating system and every major web browser when directed by a user to do so.”.

The company has taken steps to ensure that the technology is accessible to a limited group of software providers through an initiative known as Project Glasswing, to enable those companies to test and protect their own systems from potential cyberattacks.

. The unreported unauthorized access underscores the difficulty Anthropic has in completely stopping the spread of its most potent—and potentially hazardous—technology outside of authorized partners. It also begs the question of whether and for what reason anyone else might be using Mythos without authorization.

The users used a variety of strategies to enter Mythos; these included attempting popular internet sleuthing and utilizing the access as an employee of a third-party contractor for Anthropic.