Anthropic's Mythos preview: why the human layer matters more, not less
Briefly

Anthropic's Mythos preview: why the human layer matters more, not less
"Anthropic's Mythos Preview has discovered thousands of high-severity zero-day vulnerabilities, including a 27-year-old TCP SACK bug in OpenBSD and a 16-year-old flaw in FFmpeg's H.264 codec."
"The model achieves a 72.4% success rate in autonomous exploit chaining, significantly surpassing the near-zero success rate of its predecessor, Opus 4.6."
"Anthropic is launching Project Glasswing, a coalition of over 40 organizations, to use Mythos Preview defensively for finding and patching vulnerabilities in critical infrastructure."
"Experts have expressed skepticism about the claims made regarding Mythos Preview, emphasizing the need for transparency regarding false-positive rates and human-review methodologies."
Anthropic's Mythos Preview has identified thousands of high-severity zero-day vulnerabilities across major operating systems and web browsers. Notable findings include a 27-year-old TCP SACK bug in OpenBSD and a 17-year-old RCE vulnerability in FreeBSD. The model's significant advancement lies in its ability to autonomously chain exploits at scale, achieving a 72.4% success rate, surpassing skilled human researchers. Anthropic is not releasing the model publicly but has initiated Project Glasswing, collaborating with over 40 organizations to use the model defensively. Concerns about the model's claims have been raised by experts regarding false-positive rates and review methodologies.
Read at Techzine Global
Unable to calculate read time
[
|
]