What happened
On 17 July 2026, AISI published 'How Far Behind the Frontier are Leading Open Weight Models on Cyber?' — its first public quantitative measurement comparing offensive cyber capabilities of leading open-weight models (e.g. GLM-5.2, DeepSeek V4-Pro) against closed-weight frontier models. The report found open-weight models now trail the closed frontier by only 4-7 months, down from 6-10 months through most of 2025 — a narrowing gap with direct implications for defenders' planning horizons.
Why it matters
This is an authoritative capability-tracking methodology and dataset from a national AI safety/security institute, directly informing how organizations and policymakers should model the threat timeline from freely downloadable models. It follows the same pattern as NIST CAISI's GLM-5.2 assessment (already covered) and signals that risk-management frameworks referencing 'the gap to frontier' need updating faster than previously assumed.
Action needed
Security teams and policymakers tracking dual-use AI risk should incorporate AISI's updated capability-gap timeline into risk models and threat-informed defense planning; treat open-weight model releases as approaching near-frontier offensive capability on a shortening horizon.