Anthropic expands Project Glasswing to prepare cyberdefenders for powerful AI models

Anthropic expands Project Glasswing to prepare cyberdefenders for powerful AI models

Anthropic’s page, published June 2, 2026, says it is expanding Project Glasswing, its effort to secure important software, to approximately 150 new organizations. Anthropic says the group includes providers of power, water, healthcare, communications and hardware services, as well as vendors whose code is used by many organizations. It estimates that a major attack on most partners could affect more than 100 million people, with implications for national and global security. Anthropic argues that the program must do more than find vulnerabilities: the industry needs to verify, disclose and patch them, and adapt to AI models that can aid attackers as well as defenders. It says partners are using Mythos Preview to write patches and check software before release, while Anthropic is working on wider support for patching and on tools for trusted security teams. Anthropic expects many other AI companies to have Mythos-class models within 6 to 12 months, potentially without safeguards against misuse. It says robust safeguards for general access have yet to be developed, and calls their design a major challenge. Anthropic’s longer-term goal is a lasting advantage for defenders, but whether safeguards and patching can keep pace with increasingly capable models remains unknown.

Publication: 2dadde85-087c-9397-8726-06a816d73ee5

attribution — pass: The reading consistently attributes the expansion, estimates, expectations and hopes to Anthropic. The source is a single Anthropic announcement page, not a listing of other pieces.

We’re now expanding Project Glasswing.

summary — uncertain: The summary uses 'Mythos Preview' and 'Mythos-class models' as central terms but never says what they are: Anthropic's AI model, Claude Mythos Preview, which partners use to scan code for vulnerabilities. No definitions are supplied. A non-specialist cannot tell what partners are using or what other companies are expected to build. Smallest repair: on first mention, call it Anthropic's Claude Mythos Preview AI model and note that partners use it to scan code for security flaws. The rest is faithful and keeps the source's warning tone.

roughly 50 initial partners had access to Claude Mythos Preview, and since then, they’ve been deploying the model to scan their codebases for vulnerabilities

date — pass: June 2, 2026 matches the UTC day of evidence.originalPublished. The page was edited after publication (modified 2026-10-19). However, the cited passages about the expansion, the forecast and the safeguards are core announcement text, not plainly later additions. The 6-to-12-month forecast is reported as Anthropic's expectation, not an outcome. There is no hindsight framing, and this is a publisher page, not an archive copy.

within 6 to 12 months, we expect that many other AI companies will have Mythos-class models

support — pass: Each claim matches a verbatim excerpt from the source. The summary's statements about patches, pre-release checks, tools for trusted teams and a lasting advantage for defenders are supported, and they are framed as Anthropic's statements or hopes. The edits after publication do not appear to touch these core passages.

Many of Project Glasswing’s partners now use the model to write patches, as well as for pre-release checks that prevent vulnerabilities from appearing in the first place.

media — pass: No media was selected, and no verified media candidates were supplied. That is appropriate.