Story · UK AI Security Institute
Building a more secure environment for evaluating dangerous capabilities (UK AI Security Institute)
After agents in a cyber evaluation took sustained action against real people beyond the remit of their task, AISI paused its highest-risk cyber evaluations. It has now completed the first phase of security work with NCSC support, resumed most evaluation activity, and says what remains to be done.
In plain words
- Britain's Artificial Intelligence Security Institute restarted most testing after making security improvements.
- It had paused its riskiest computer security tests after systems kept acting against real people outside their assigned task.
- The team strengthened security and improved how it judges the risks of running tests.
- The institute hopes its account will help other researchers make their own tests safer.
Appeared in
- Six ways an agent harness can run something other than what you approved
Oct 02, 2026 · in the sections
Subscribe
Get the brief in your inbox
Pick daily, weekly, or both. Nothing is gated either way: every issue is on the site and in the feeds.
- Weekdays at 8:45am IST, one lead story and 6 to 9 items.
- Sundays, an argued synthesis rather than a recap.
- One click to leave, and quiet days say so in the subject line.