Anthropic is expanding Project Glasswing. The move gives more trusted security partners access to Claude Mythos Preview. Anthropic wants the model to help defenders before similar tools become easier to reach.
The new wave adds roughly 150 organizations across more than 15 countries. Each partner still has to meet Anthropic’s security requirements before it can use the model.
That is a major jump from the first cohort. In April, Anthropic said about 50 partners were using Claude Mythos Preview to scan codebases. The company later said those partners had found more than 10,000 high- or critical-severity flaws. We covered that earlier in our Claude Mythos Project Glasswing report.
Why Anthropic is widening Project Glasswing
The expansion focuses on sectors where one breach could ripple far beyond one company. Anthropic named power, water, healthcare, communications, hardware, and software vendors. Many of those vendors maintain code that other groups and governments rely on.
In Anthropic’s framing, the stakes are national and global. For most of the incoming partners, the company estimates that one major attack could affect more than 100 million people.
The company is also preparing defenders for stronger cyber-capable models. Anthropic warns that other labs may have Mythos-class systems within 6 to 12 months. Those releases may not arrive with the same guardrails.
Access is still limited
Claude Mythos Preview is not becoming a normal public Claude model yet. Anthropic says it wants broader Mythos-level access. However, it still needs safeguards that can block malicious cyber use without slowing legitimate defense work.
For now, Project Glasswing remains the controlled path. Anthropic says partners use Mythos Preview to find vulnerabilities, write patches, and run pre-release checks. They can also test systems, automate threat response, and explore safer ways to rebuild old codebases.
Anthropic also points to its technical assessment of Claude Mythos Preview as part of the caution. The model’s cyber abilities can help defenders. That same class of ability could also make attacks faster and harder to predict.
The big takeaway is simple. Anthropic is not just showing off another frontier model benchmark. It is building an operating model for what happens when AI changes the economics of cybersecurity.
