Anthropic expands Project Glasswing to 150 new organizations with $100M in Claude Mythos Preview credits after partners surface 10,000+ critical vulnerabilities
Anthropic extended Project Glasswing this week from the original twelve launch partners to approximately 150 additional organizations, committing up to $100M in Claude Mythos Preview usage credits and $4M in direct donations to open-source security maintainers. The first-cohort partners — AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks — surfaced more than 10,000 high- or critical-severity vulnerabilities in widely-deployed software during the initial six-week run.
The expansion is the operational answer to a hard question: if a frontier model has reached coding capability where it can find and exploit vulnerabilities better than nearly all human researchers, what is the responsible release path? Anthropic's answer is not "ship it" and not "sit on it" — it is a controlled-access partner program where defenders get the capability first, in volume, before the model goes broad. The thousands of zero-days surfaced across operating systems, browsers, and infrastructure software during the initial phase are the evidence that the capability is real and that the patch-first sequencing matters.
The $100M credit commitment is the part of the announcement that signals scale of intended use rather than scale of compute generosity. Mythos Preview inference is expensive; $100M routed through 150 organizations is enough to run sustained, repeated security audits across every critical-infrastructure codebase those organizations maintain. The $4M to open-source security organizations addresses the gap the partner list otherwise leaves — the maintainers of the upstream packages every Glasswing partner depends on but does not staff.
The throughline back to broader frontier-model policy: Anthropic has stated it will not make Mythos-class capability widely available until it has implemented "highly robust safeguards" against misuse. That positioning is consistent with the company's longstanding responsible-scaling framing and inconsistent with the model-release cadence the open-weight cohort has set. Whether the controlled-access window measured in weeks for Glasswing turns into months — or longer — for public Mythos availability is the release-timing question that will define Anthropic's H2 2026.
Anthropic — Expanding Project Glasswing → · Engadget — Anthropic expands its Claude Mythos preview to more partners → · Gizmodo — Anthropic Lets Claude Mythos Spread Its Glasswings →