What AI can now do·Apr 7, 2026·Cyber

Claude Mythos Preview reports thousands of critical software flaws; release withheld

Anthropic

17 hoursMETR 50% time horizon (early snapshot)

Anthropic said Mythos Preview could find and exploit previously unknown flaws across every major operating system and browser, reported thousands of high- and critical-severity findings (99% still unpatched at announcement), scored 93.9% on SWE-bench Verified and measured about a 17-hour METR horizon; it withheld general release and gave access only to vetted defenders through Project Glasswing.

How far it has travelled

From a demonstration to real life, in five steps. Empty steps are the point: a demonstration is not a deployment.

Demonstratedshown once, by its developer or a researcher
yes · Apr 7, 2026 · Thousands of high- and critical-severity findings (89% of 198 manually reviewed reports matched its severity rating); it completed a corporate-network cyber range estimated at 10+ expert hours that no prior model had; METR 1.1 horizon 1,044.8 min (CI 509-3,304). source
Reproducedshown again by someone else
yes · Sep 3, 2026 · OpenAI treated GPT-5.3-Codex as 'High' for cyber (Feb 2026) and rated GPT-6 Astra 'Critical', the top tier (Sept 2026); Google could not rule out Gemini 3.1 Pro's cyber alert threshold (Feb 2026) or Gemini 4 Argon's cyber CCL (Oct 2026). source
Accessibleavailable to the public or to customers
partial · Oct 6, 2026 · No general release: 11 launch partners plus 40+ organizations (up to $100M in credits; $25/$125 per million tokens after the preview), about 150 organizations by June 2; Fable 5 (same model, cyber and bio requests routed to Opus 4.8) was released generally on June 9 at $10/$50; a three-tier Cyber Verification Program with Mythos 5.1 from Oct 6. source
Reliableworks most of the time on a defined task
partial · Apr 22, 2026 · Mozilla fixed 271 Firefox bugs identified in one initial evaluation; critics note the browser tests used a reduced-security build, almost every success relied on two now-patched bugs, and one headline bug was also found by 8 of 8 open models tested by AISLE; 'thousands' is Anthropic's extrapolation. source
Real consequenceshas changed something in the world, for good or ill
yes · Jun 30, 2026 · Treasury, the Fed and foreign central banks convened banks within hours; Anthropic delayed general Mythos-class release pending new cyber safeguards (May 28: 'coming weeks'; Fable 5 on June 9); the US barred non-US access June 12-30; outsiders reached the withheld model via a vendor; earlier snapshots left their sandbox in tests; Claude models reached three real organizations during evaluations. source

What this could change for you

The software behind banks, phones and hospitals is being audited by machines faster than it can be fixed, so expect more emergency updates and precautionary outages.

Incidents it made possible

Sources

primary source, second source.

A milestone is a dated, sourced result that moved what AI can do; a stage is upgraded only when a source shows it (a reproduction, a public release, production evidence, a documented consequence). “Unknown” is the honest default.