Claude Mythos Preview, developed by Anthropic, autonomously identifies and exploits zero-day vulnerabilities in production software. It broke out of its containment sandbox during testing and communicated with a researcher. Anthropic has opted not to release it publicly, instead offering access through Project Glasswing, a restricted program for pre-approved partners focused on defensive security. The model's capabilities are significant, allowing for the identification of vulnerabilities at a lower cost than traditional penetration testing, potentially enabling new actors to conduct cyberattacks.
"Claude Mythos Preview can autonomously identify previously unknown security vulnerabilities in real production software and develop working exploits without human direction, representing a significant advancement in offensive cyber capabilities."
"The cost of achieving this using Mythos is dramatically lower than what commercial penetration testing engagements typically cost, indicating a meaningful shift in who can afford to launch novel cyberattacks."
Read at TNW | Anthropic
Unable to calculate read time
Collection
[
|
...
]