Anime, manga, and games, with a take · A Yukimedia publication

Anthropic's Project Glasswing Logs 629 Confirmed Fixes Against 37,966 Candidates

Anthropic's Project Glasswing, started April 7, 2026, uses the Claude Mythos Preview model to find software vulnerabilities and route them to developers. By a November 2 tally, Claude models had produced 37,966 candidates since November 1, 2025.

Reporting from 1 source: GIGAZINE.

Why it matters

The gap between 5,087 reports and 629 confirmed fixes is the number that matters: finding candidates with AI has scaled faster than the disclosure and patching work behind them.

Anthropic's Project Glasswing Logs 629 Confirmed Fixes Against 37,966 Candidates

Key Facts

  • Anthropic began Project Glasswing on April 7, 2026, providing Claude Mythos Preview to organizations including Microsoft, Apple, and the Linux Foundation.
  • Claude models produced 37,966 vulnerability candidates between November 1, 2025 and the November 2, 2026 tally.
  • External security companies verified 6,376 candidates and judged 6,009 of them to be vulnerabilities.
  • Reports sent to developers rose from 1,596 on May 22, 2026 to 5,087 on November 2, 2026.
  • Anthropic confirmed published fixes for 629 reports as of November 2026, up from 97 in May 2026.

External firms verified 6,376 and judged 6,009 to be vulnerabilities. Reports to developers reached 5,087, and Anthropic confirmed published fixes for 629.

VulnCheck's Patrick Garrity examined Anthropic's public ledger and found 7,349 entries as of November 2, 2026, including candidates still at the pre-report stage where only a hash and a date are published. The ledger's own "Fixed" count was 217, well below the 629 fixes Anthropic says it has confirmed across all reports.

Anthropic's dashboard also splits the 5,087 reports: 1,862 went through outside verification, while 3,225 were sent to developers directly. Garrity counted roughly 12,000 new candidates between September 2026 and November 2, against a much smaller rise in records marked fixed.

Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.

Add Yomimono as a preferred source on Google

Sources