Anthropic Expands AI Security Push

Anthropic is broadening its ambitious Project Glasswing initiative, a program focused on securing critical software. Initially launched with around 50 partners testing the AI model Claude Mythos Preview for vulnerability detection, the project is now extending its reach.

The program is onboarding roughly 150 new organizations, many of which operate in sectors like power, water, healthcare, communications, and hardware. These partners, spanning over 15 countries, must meet stringent security requirements before gaining access.

A key focus of the Anthropic Project Glasswing expansion is to onboard vendors whose codebases are widely relied upon globally, including by governments. The company estimates that a successful attack on these partners' systems could impact over 100 million people.

Shifting Cybersecurity Norms

Anthropic sees Project Glasswing as a crucial step toward making all software more secure through AI. The expansion reflects growing industry and government conversations about AI's transformative role in cybersecurity.

The company anticipates that powerful AI models, potentially comparable to Mythos, will become widely available within months, possibly without robust safety safeguards. This underscores the urgency for defenders to adapt.

Anthropic's strategy involves providing broad access to advanced AI models and tools while helping the industry shift from vulnerability discovery to more efficient patching and disclosure processes.

Accelerating Patching and Defense

The bottleneck in cybersecurity is increasingly the verification, disclosure, and patching of vulnerabilities that advanced AI can uncover. Mythos Preview is already being used by partners to write patches and perform pre-release checks.

Anthropic is also releasing tools to trusted security teams and developing initiatives to scale up vulnerability patching, particularly for open-source software. This includes improving the process for disclosing vulnerabilities to maintainers.

Looking ahead, Anthropic plans further expansions of Project Glasswing, prioritizing essential infrastructure providers and maintainers of critical open-source software. The company is working on safely releasing Mythos-level capabilities more broadly, acknowledging the significant challenge of developing precise and robust safeguards against misuse.