Anthropic launches Glasswing

Computer security is the work of finding software flaws before attackers do, and Anthropic says its new Project Glasswing is built for that job. (anthropic.com) Anthropic announced the program in April 2026 and said launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. (anthropic.com) The companies will use Claude Mythos Preview, a restricted Anthropic model, to scan first-party and open-source software for vulnerabilities and share lessons Anthropic says it will publish for the wider industry. (anthropic.com) A vulnerability is a bug that can be turned into a break-in, and a zero-day is one that defenders do not know about yet. Anthropic said Mythos Preview can identify and exploit zero-day vulnerabilities in every major operating system and every major web browser in its testing. (red.anthropic.com) Anthropic said more than 99% of the vulnerabilities it found with the model were still unpatched, which is why it withheld technical details and said it is following coordinated disclosure practices. (red.anthropic.com) The company is not releasing Mythos Preview broadly. Anthropic said it plans to keep access limited because the same capability that helps defenders patch software could also help attackers reverse-engineer exploits and weaponize known flaws faster. (red.anthropic.com) Amazon Web Services said Claude Mythos Preview is available in gated research preview through Amazon Bedrock as part of Project Glasswing, describing the model as able to comprehend large codebases and produce actionable security findings with less manual guidance than earlier systems. (aws.amazon.com) Anthropic tied the launch to a broader shift in artificial intelligence safety policy. Its Mythos Preview system card says the model marks a large jump over Claude Opus 4.6 on benchmarks and includes detailed safety evaluations because the company sees cybersecurity as one of the areas where capability gains can create immediate misuse risk. (anthropic.com) The project puts large platform companies and security vendors on the same side of the table: the firms that build operating systems, cloud infrastructure, browsers, chips, and endpoint defenses are testing one model against the software stacks they already maintain. (anthropic.com) Anthropic said it has also extended access to more than 40 additional organizations that build or maintain critical software infrastructure. For now, Glasswing is less a public product than a closed early-warning network built around a model Anthropic says is already good enough to find bugs humans miss. (anthropic.com)