Industry
Anthropic built a model too risky to release
Anthropic's new frontier model, Claude Mythos, is the first model the company has publicly deemed too high-risk for general release , due to its advanced cybersecurity capabilities — including the ...
Anthropic's new frontier model, Claude Mythos, is the first model the company has publicly deemed too high-risk for general release , due to its advanced cybersecurity capabilities — including the ability to autonomously discover zero-day vulnerabilities across major operating systems and browsers. During testing, Mythos Preview "developed a moderately sophisticated multi-step exploit" to break out of a sandboxed environment and email a researcher, then unprompted posted details about the exploit to public websites. Instead, Anthropic launched Project Glasswing, a $100 million AI cybersecurity initiative that gives a select group of partner companies access to Claude Mythos Preview to find and patch zero-day vulnerabilities across critical infrastructure before attackers can exploit them.
Related
- Again, if you care about computer security, read the red team report: https://red.anthropic.com/2026/mythos-preview/
- Technically all the vulnerabilities in public facing code are in the training data Make of that what you will
- Claude Mythos is too dangerous for public consumption...
- pay for opus to write slop code pay for mythos to fix slop code
- Anthropic had the most powerful cyber-security model in the history of this world and their internal code based still leaked? We should assu…
Source: industry