Anthropic is investigating 'unauthorized access' of its Mythos cybersecurity tool
Engadget • • business
Key Points:
- Anthropic is investigating unauthorized access to its Claude Mythos AI model, which is designed to identify cybersecurity vulnerabilities, after a group gained entry via a third-party contractor portal.
- The unauthorized group reportedly used internet sleuthing techniques and accessed the model through a developer portal by guessing its location, but sources say they are interested only in experimenting, not malicious use.
- Claude Mythos Preview was launched recently under "Project Glasswing" and is currently restricted to select trusted companies like Amazon, Microsoft, Apple, Cisco, and Mozilla, which used it to identify and fix 271 Firefox vulnerabilities.
- Banks and government agencies have shown growing interest in the model for enhancing their cybersecurity defenses.
- The same group may also have accessed other unreleased Anthropic models, raising concerns about broader security implications.
