The AI Security Institute on Require AI labs to publish safety evalu…

The AI Security Institute

UK government research organisation focused on AI safety and security

Evaluations should report capability curves, especially when performance may still be rising.

AI Verified (Jul 2, 2026)

Source:

The AI Security Institute, “More compute, more capability: Why AI agent evaluations need to account for test-time compute,” July 2, 2026.

2mo ago

Policy proposals and claims

Require AI labs to publish safety evaluations before deploying frontier models

votes For

Statement relation verification history AI Unverifiable Report this

AI Unverifiable The quote favors reporting evaluation curves, but the cited context does not establish a stance on publishing safety evaluations before deployment. · Hector Perez Arenas gpt-5 · 2mo ago

Vote inference verification history Unverified Report this

No vote answer verification comments yet.

Quote authenticity verification history

Report this

AI Verified Verified against the cited source dated 2026-07-02; wording and attribution match. · Hector Perez Arenas gpt-5 · 2mo ago

replying to The AI Security Institute

Other opinions from The AI Security Institute

See all

The AI Security Institute votes For

Governments should fund open-source AI safety tools and red-teaming infrastructure as public goods.

By open-sourcing not only Inspect, but also our methods and infrastructure where we can, we hope to help researchers, organisations, and other evaluators stand up rigorous evaluation capability without starting from zero. We hope that the Engineering...

more AI Verified source (Jun 18, 2026)

added 2mo ago

The AI Security Institute votes Against

AI alignment is solvable

A more fundamental fix would be to train the models not to cheat in the first place – but given this kind of behaviour was reported in frontier models more than a year ago, robustly aligning it away may not be easy.

AI Verified source (Apr 27, 2026)

added 1mo ago