UK AI Safety Institute releases Inspect

Open evaluation software

The institute made Inspect available to the wider community on Friday, May 10, 2024. The software supports model evaluations and scoring across areas including knowledge, reasoning and autonomous capabilities. Its code was released under the MIT license, as confirmed by a repository snapshot from the launch period. [1][2]

The government’s announcement explicitly says available that day, and May 11 reporting describes the Friday release. The institute was called the AI Safety Institute at launch; its current Atlas organization name reflects the later Security Institute name. Releasing an evaluation framework does not certify the safety of every model tested with it. [1][3]