UK AISI says leading open-weight models are now only 4–7 months behind frontier models in cyber ability
新智元 · wechat · 2026-07-21
The UK’s AISI published a public report quantifying the cyber gap between leading open-weight and closed frontier models for the first time: roughly 4–7 months behind, down from 6–10 months in last year’s internal tests.
- AISI used two evaluation tracks: 70 narrow cyber tasks and a multi-step CyberRange simulating enterprise intrusion chains.
- GLM-5.2 roughly matched Opus 4.6 on narrow tasks and Opus 4.5 on CyberRange; DeepSeekV4-Pro tracked close behind on both.
- The cost gap is even larger than the capability gap: on the same CyberRange test, Opus 4.5/4.6 cost about $85 per run, GLM-5.2 about $46, and DeepSeekV4-Pro just $1.19.
- AISI argues that once open weights are released, the capability is permanently out in the wild, while defensive tools still require each team to invest time and money.
- The report’s policy implication is sharper than the raw numbers: decision-makers may need to define which capability levels should never be released as open weights.
More from AGI Musings
- AI folks, stop saying "biology" when you mean drug development — KevinKaichuang · 2026-10-26
- Anthropic: junior tech hires fell from 27% to 9% as agent hours hit 6x employee hours — thione · 2026-10-26
- Anthropic's Jascha: you may have only 1–5 years of intellectual contribution before AGI — GabGarrett · 2026-10-26
- Anthropic's S-1 Shows $6.6M Revenue Per Employee as AI Agents Log 6x Human Hours — jacalulu · 2026-10-26
- Ben Recht: The world needs a truly open language model, and it's easier to build than you think — beenwrekt · 2026-10-26
- Which jobs fall to AI first? The rule: any layer whose rules can be written as code — FinanceYF5 · 2026-10-26