Organization
What is uk aisi?
UK AISI (AI Safety Institute) is a UK government body focused on evaluating frontier AI models for safety, particularly through jailbreaking tests.
Company timeline
- Mar 2026 - Geoffrey Irving said UK AISI had never failed to jailbreak a model, noting that strong jailbreaks are concentrated in domains like bio.
- Mar 2026 - Geoffrey Irving stated UK AISI is shifting work to longer research collaborations, both post-deployment and earlier before deployment is finalized, due to increasing pace of model releases.
Where it appears in the record
Every line below is attributed to a named speaker.
Worth quoting
Geoffrey Irving on UK AISI jailbreak testing: every model tested was broken.
“Every time we did we jailbroke a model.”Geoffrey Irving · 1 Mar 2026
Worth quoting
UK AISI chief scientist asserts that no AI model has ever been made robust against jailbreaking.
“We've never failed to jailbreak a model. None of them are robust.”
By the numbers
UK AISI is shifting from pre-deployment evaluations to longer post-deployment research collaborations due to an accelerating pace of model releases.
“We are shifting a lot of that work not least because the pace of model releases is increasing to longer research collaborations that might go back either are post deployment or they go back further before deployment is finalized.”Geoffrey Irving · 1 Mar 2026
By the numbers
UK AISI found that the strongest jailbreaks are concentrated in biosecurity and cybersecurity domains specifically.
“The strong jailbreaks are concentrated in very particular domains. And often that list of domains is bio.”Geoffrey Irving · 1 Mar 2026
Citation Bureau · reference note, compiled from attributed expert discussion. Last updated 2026-06-14.