AI BriefingMythos Leak
Key point
Anthropic's Claude Mythos Preview was reportedly accessed without authorization.
Details
According to Bloomberg reports, Claude Mythos Preview was accessed without authorization by a small group of Discord users shortly after its release. One member of the group was reportedly a third-party contractor for Anthropic.
They reportedly used operational information leaked from an earlier Mercor breach to infer where the model was located, and are said to still maintain access and be using it.
Anthropic stated it is investigating reports of unauthorized access through a third-party vendor environment. Prior to release, Mythos was distributed on a limited basis to only 40 companies, including Microsoft, Apple, and Google.
Mythos is a model touted for its security vulnerability detection performance, and Anthropic stated it used the model to find a 27-year-old vulnerability in OpenBSD, while Mozilla said it identified and fixed 271 vulnerabilities in Firefox.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Anthropic Blocks 'Malicious Use' of AI Capable of Developing Biological Weapons