In April 2026 Anthropic said it would not generally release Claude Mythos Preview, a model it said had found thousands of serious vulnerabilities in major operating systems and browsers, and instead gave access to about 50 companies and organisations that maintain critical software.

Evidence1 sourceCited by 1 pointCreated Nov 5, 2026

Source

VentureBeat: Anthropic says its most powerful AI cyber model is too dangerous to release
Published by venturebeat.comRecorded Nov 5, 2026
Open the original

Supports: Anthropic kept its most dangerous model from the public and let defenders patch software flaws first.
- A reason for “Anthropic”