Claude Mythos's Exclusivity Challenged by Open Models
2 sources·2 updates
Recent studies have shown that smaller open models can replicate the cybersecurity findings of Anthropic's Claude Mythos, undermining its perceived exclusivity. Notably, the AISLE tests identified 15 vulnerabilities in OpenSSL and five in curl, while all tested models flagged a critical FreeBSD bug.
Key Points
All eight models tested, including GPT-OSS-20b, identified the FreeBSD bug.
Only GPT-OSS-120b recognized the patched version of the FreeBSD code as safe.
Timeline
Get personalized news summaries delivered to your feed
Try Trace Free