UK AI Security Institute: open-weight AI now trails the closed cyber frontier by 4-7 months
The UK AI Security Institute found open-weight GLM-5.2 and DeepSeek V4-Pro now trail closed frontier cyber capability by 4-7 months, down from 6-10 months in 2025, with DeepSeek clearing a cyber-range task at $0.28 versus $12.50 for Opus 4.5. With open-model guardrails bypassed by retrying a blocked request, frontier-grade offensive cyber capability is moving toward cheap, unmonitored reach — narrowing the window defenders have to prepare before it spreads without safeguards.
Source: aisi.gov.uk ↗
Once open weight models are released, these options are lost permanently: safeguards can be removed, and copies can be downloaded, redistributed, and run on private systems beyond monitoring.
UK AI Security Institute
Why this matters
- → Open-weight AI cyber capabilities closing 4-7 months to frontier, down from 6-10 months.
- → Guardrails easily bypassed via retry; dangerous capabilities spreading to unmonitored systems.
- → Defenders' preparation window shrinking as cheap, accessible cyber tools proliferate.
The gap narrows