Anthropic: Claude Mythos Found 10,000+ Critical Bugs in a Month
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic said Claude Mythos Preview identified more than 10,000 high- or critical-severity vulnerabilities within roughly a month of Project Glasswing's launch, a collaborative effort to secure critical software before capable AI models are turned against it
- The Decoder reports Mythos Preview is finding bugs faster than developers can patch them, and Anthropic's own post warns the software industry must adapt to the volume of vulnerabilities such models will surface
- Project Glasswing is expanding to additional partners per Benzinga, with participating software companies releasing "multiple times as many patches as usual" according to LinkedIn commentary from those companies
- Ubiquiti patched three max-severity UniFi OS vulnerabilities per BleepingComputer, a concrete downstream example of the patching pressure the initiative is generating
- Box CEO Aaron Levie framed the shift succinctly: the new bottleneck is no longer finding security issues but reviewing, responding to, and fixing them
- Anthropic published a coordinated vulnerability disclosure dashboard and new LLM exploit-evaluation results alongside the update, while team member Newton Cheng flagged exploitation capability as a "real leap" over previous models
Why it matters: Roughly 10,000 high- or critical-severity findings in one month shifts the cybersecurity bottleneck from detection to response — defenders' triage, patching, and coordinated-disclosure pipelines must now scale to match what Mythos Preview surfaces, or face an expanding backlog of publicly known vulnerabilities that attackers can race to weaponize.
Ask SkimNews
