Anthropic launches Project Glasswing
Anthropic announced Project Glasswing, an AI cybersecurity initiative that the company says will use its Claude Mythos model and include partners such as Nvidia, Microsoft, Cisco and JPMorgan Chase. The project is presented as an attempt to apply frontier AI to cybersecurity use cases with a coalition of large vendors and enterprises. (telecompaper.com)
Anthropic said on April 7 that it is launching Project Glasswing, a cybersecurity program that gives a small group of companies early access to its Claude Mythos Preview model. (anthropic.com) The launch group includes Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks, according to Anthropic’s announcement. (anthropic.com) Anthropic said the project is aimed at “the world’s most critical software,” with partners using Claude Mythos Preview for defensive security work and Anthropic sharing lessons from the program with the wider industry. (anthropic.com) The company said it is committing up to $100 million in usage credits and $4 million in donations to open-source security groups, while extending access to more than 40 additional organizations that build or maintain critical software infrastructure. (anthropic.com) Cybersecurity here means finding weak spots in code before attackers do. Anthropic’s security team said Mythos Preview can discover “zero-day” flaws — bugs that are still unknown to defenders — in real open-source codebases and can also turn some already-known software flaws into working exploits. (anthropic.com) Anthropic said more than 99% of the vulnerabilities it found with Mythos Preview have not yet been patched, which is why it withheld technical details and said it is following coordinated vulnerability disclosure practices. (anthropic.com) The model itself is not being released broadly. Anthropic said in its Mythos Preview write-up that it does not plan to make the system generally available, even as it gives limited research access to selected outside users and deploys it internally for security and other work. (anthropic.com; anthropic.com) Anthropic’s system card for Mythos Preview says the model shows a “striking leap” over its previous frontier model, Claude Opus 4.6, on several benchmarks, including computer security evaluations. (anthropic.com) Microsoft, Cisco and NVIDIA are already tied into Anthropic through separate infrastructure and security partnerships. Microsoft said in November 2025 that Anthropic would scale Claude on Microsoft Azure with NVIDIA support, while Cisco has spent the past year expanding its Secure AI Factory work with NVIDIA. (blogs.microsoft.com; newsroom.cisco.com) For Anthropic, Glasswing turns that network into a test bed: large cloud providers, hardware makers, software vendors and one of Wall Street’s biggest banks all using the same model to hunt for software weaknesses before attackers automate the same work. (anthropic.com; anthropic.com)
