Anthropic says it intends to eventually make Mythos-class vulnerability-finding artificial intelligence available more broadly, but for now is expanding its restricted Project Glasswing program to additional partners including U.S. and allied governments. The company says Mythos has scanned more than 1,000 open-source projects, estimated 6,202 high-or-critical-severity vulnerabilities and 23,019 total flaws, and validated many findings through coordinated disclosure; no CVE list or release date for public access was provided.
Mayank Parmar
2026.06.10
89% relevant
This article advances the same underlying event by reporting Anthropic's public rollout of Fable 5, a guarded version of the Mythos-class model, and adds concrete details on access limits, sensitive-query downgrading to Opus 4.8, temporary availability to Pro/Max/Enterprise users, and the distinction between restricted Mythos 5 and safeguarded Fable 5.
Eduard Kovacs
2026.06.09
84% relevant
This article is a direct follow-up on that same underlying event: Anthropic has now launched Claude Fable 5 for general availability with cyber/bio fallbacks and says Project Glasswing partners are being upgraded from Mythos Preview to Mythos 5, adding concrete rollout details, guardrail design, pricing, and partner-access changes.
Ionut Arghire
2026.06.09
93% relevant
This is a direct follow-up on the same Mythos program, adding concrete exploit-generation results: Anthropic says Mythos Preview produced working Firefox and Windows N-day exploits within hours, showing the model can weaponize disclosed flaws rather than only find bugs.
2026.06.03
86% relevant
This article adds specific details on Anthropic's Project Glasswing expansion from about 50 to 200 partners across 15 countries, identifies new access dynamics affecting UK banks, and notes that ENISA will receive Mythos Preview access while CISA has not yet been selected.
2026.06.02
88% relevant
This article directly updates the same Mythos / Project Glasswing event by adding that Anthropic expanded the preview program by about 150 organizations to roughly 200 total partners, and that Cisco used Mythos Preview and OpenAI's GPT 5.5-Cyber to scan 1.8 billion lines of Cisco code in eight weeks with a reported false-positive rate under 3 percent, though Cisco did not disclose the number of flaws found or fixed.
Eduard Kovacs
2026.06.02
96% relevant
This is a direct update on the same underlying event: Anthropic broadening Mythos availability. The new reporting adds that Project Glasswing is expanding from about 50 to roughly 200 total partner organizations, that the new cohort includes critical-infrastructure entities and reportedly Okta, Samsung, ENISA, and NATO, and that Anthropic says Mythos has found more than 23,000 potential vulnerabilities with only 75 high/critical issues patched so far.
Mayank Parmar
2026.05.29
95% relevant
This article directly updates the same underlying event by reporting Anthropic's confirmation that Mythos-class models are expected to roll out to all customers in the coming weeks, adding timing and reaffirming that prior restrictions were due to security risk concerns.
Mayank Parmar
2026.05.25
93% relevant
This article adds specific evidence that Anthropic may be moving from restricted access toward product integration, citing Mythos references and a briefly exposed toggle in public Claude Code and Claude Security, plus new details on the Glasswing program and Anthropic's claim that Mythos found 10,000 high- or critical-severity vulnerabilities in its first month.
Eduard Kovacs
2026.05.25
91% relevant
This is a direct update on the same underlying Mythos/Project Glasswing rollout, adding Anthropic's first large-scale outcome data: 23,000 potential flaws across 1,000+ open source projects, 1,726 externally confirmed findings, more than 1,000 high- or critical-severity issues, 75 severe issues already patched, and 65 advisories published.
2026.05.25
100% relevant
This article establishes a new trackable development around Anthropic’s Mythos program by adding a concrete policy shift toward wider access, naming government expansion, and quantifying validated open-source vulnerability discovery at scale.