AI Security Institute
A UK government body that studies and tests the safety of advanced AI systems.
The AI Security Institute (AISI) is a body set up by the UK government to study, test and advise on the safety of advanced AI systems. It carries out independent evaluations of leading models, often in cooperation with the labs that build them.
AISI’s work matters because it brings rigour to claims about what AI can and cannot do. In one study, for example, it found that the standard way of testing AI agents quietly understated their abilities — because the tests cut the agents off before they had enough computing time to finish. Given more time, the same models solved far more.
As governments move to govern AI, institutes like AISI are becoming an important, relatively neutral source of evidence in a field crowded with marketing.