What AI can now do·Nov 1, 2024·Cyber

Big Sleep finds a previously unknown SQLite flaw, a first for AI agents

Google Project Zero / Google DeepMind

about 150 CPU-hoursFuzzing that failed to find the bug

Google's Big Sleep agent (Gemini 1.5 Pro) found a previously unknown, exploitable security flaw in the SQLite database engine that about 150 CPU-hours of fuzzing had missed; it was reported in early October 2024 and fixed the same day, before any release shipped it.

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 · Nov 1, 2024 · Google called it the 'first public example' of an AI agent finding an unknown exploitable memory-safety issue in widely used real-world software, and described the work as 'highly experimental'. source
Reproducedshown again by someone else
yes · Apr 7, 2026 · Anthropic's Mythos Preview reported thousands of high- and critical-severity vulnerabilities, including in every major operating system and browser (Apr 2026); OpenAI's Daybreak program said its cyber-tuned model found two high-severity Chrome bugs (Aug 2026). source
Accessibleavailable to the public or to customers
partial · Oct 6, 2026 · Frontier bug-finding is gated to vetted defenders: Anthropic's Glasswing and three-tier Cyber Verification Program, OpenAI's Daybreak tiers with hardware keys from Sept 2026; general models such as Opus 4.6 are still said to 'find real vulnerabilities in codebases'. source
Reliableworks most of the time on a defined task
partial · Apr 22, 2026 · Google judged a target-specific fuzzer at least as effective as Big Sleep in 2024; by April 2026 Mozilla had fixed 271 Firefox bugs identified in one initial Mythos evaluation. source
Real consequenceshas changed something in the world, for good or ill
yes · Oct 21, 2026 · AI-found flaws in PHP's core (CVSS 9.2) led agencies in the US, UK, EU, Canada, Australia and Japan to urge patching and sites in Finland, the Netherlands, Belgium and France to go offline as a precaution; Google reported criminals using an AI-built zero-day in May 2026. source

What this could change for you

Bugs that hid in everyday software for decades are now found in days, by defenders and, inevitably, by attackers.

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.