Overview
- Anthropic said on Tuesday it expanded Project Glasswing by about 150 organisations, bringing total partners to roughly 150–200 across more than 15 countries and including government bodies and critical‑infrastructure vendors.
- Claude Mythos has scanned over 1,000 open‑source projects and flagged about 23,019 potential vulnerabilities, with the company estimating roughly 6,202 as high or critical and partners validating more than 10,000 serious flaws.
- Access remains tightly controlled because Mythos can both find zero‑day bugs and autonomously generate working exploits, so each new partner must meet security requirements before using the model.
- Security teams report the main bottleneck is human capacity to triage, verify, disclose, and deploy patches for the surge of AI‑found issues, and Anthropic says it will scale verification, offer credits and grants, and help coordinate patching.
- The expansion follows Anthropic’s confidential IPO filing and comes as rivals prepare similar cyber tools, raising pressure on regulators and prompting questions about cross‑border safeguards and how quickly Mythos‑class models will spread.