AI BriefingMythos Unauthorized Access
Key point
Anthropic's unreleased Mythos was reportedly accessed without authorization in a third-party vendor environment.
Details
Anthropic's unreleased Mythos Preview was reportedly accessed without authorization shortly after its announcement.
- According to Bloomberg, some users in a private Discord chat gained access, and one of them was an Anthropic third-party contractor.
- They reportedly inferred the model's location based on Anthropic operational information leaked in the earlier Mercor hack.
- Anthropic said it is investigating reports of unauthorized access through third-party vendor environments.
- The group reportedly still retains access and has not yet used the model in cyberattacks.
Anthropic distributed Mythos to a limited group of 40 companies, including Microsoft, Apple, and Google. The company claimed this model is stronger at vulnerability detection than previous versions, and stated it was actually used to find 1 27-year-old OpenBSD vulnerability and 271 vulnerabilities in Firefox.
Fortune had previously reported that a security incident before the public release exposed Mythos's existence.
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