AI Cyber Watch

Dual-track timeline

Capability above, incidents below

Published lab capability points on the top track, reported incidents and context on the bottom track.

Caveat: Co-occurrence, not proof of cause. Events appearing close together does not mean one caused the other. We do not calculate a correlation coefficient: there are too few points, and they measure different things.

CapabilityApprentice-level cyber task success: 9% (Late 2023)
CapabilityHCAST: 70 to 80% success on tasks taking skilled humans under one hour (Mar 2025)
ContextNCSC assessment: AI and the cyber threat to 2027 (7 May 2025)
CapabilityCyberGym: about 20% success on vulnerability reproduction tasks (Jun 2025)
CapabilityFirst expert-level (10+ years experience) cyber task completions (2025)
CapabilityFrontier Safety Framework defines tracked cyber capability levels (Sep 2025)
ContextENISA threat landscape provides EU incident context (1 Oct 2025)
ContextImpact lesson: a stopped production line (JLR, 2025) (22 Oct 2025)
CapabilityApprentice-level cyber task success: 50% (Late 2025)
Reported activityAI-orchestrated espionage campaign reported by Anthropic (13 Nov 2025)
ContextJoint guidance on integrating AI into operational technology (3 Dec 2025)
CapabilityMythos Preview: first model to complete The Last Ones, 3 of 10 attempts (Apr 2026)
CapabilityGPT-5.5 completed The Last Ones in 2 of 10 attempts (Apr 2026)
CapabilityNewer Mythos Preview checkpoint: The Last Ones 6 of 10 (May 2026)
CapabilityFirst ever completion of the Cooling Tower ICS range: 3 of 10 (May 2026)
CapabilityLab system card reported a broader cyber evaluation suite (Jul 2026)
Reported activityTaiwan government confirms hybrid AI-agent campaign reporting (13 Aug 2026)
ContextNCSC alert: increased targeting of operational technology (27 Aug 2026)
CapabilityLab system card treated the model as High cyber capability under its framework (2026 system card)
CapabilityArtificial Analysis Cyber Index leaders scored 56 (Sep 2026)
Published lab point Reported AI-orchestrated activity Context, not AI-attributed

Evidence routes

Sources for every item are on the Frontier and Incidents pages.