Inspect: A framework for large language model evaluations
Department for Science, Innovation and Technology (DSIT)
Ministerial department
https://www.gov.uk/government/organisations/department-for-science-innovation-and-technology
Total FTE: 2,275·Digital & data FTE: 130
Sub-organisations: Government Digital Service (GDS), Intellectual Property Office (IPO), Ordnance Survey (OS), UK Space Agency, UK Research and Innovation (UKRI), Met Office
Stars of active repositories
2,643
Active repositories
45
Total repositories
148
GitHub organisations
Repositories
Showing 1–10 of 45 repositories, sorted by stars
Collection of evals for Inspect AI
ControlArena is a collection of settings, model organisms and protocols - for running control experiments.
A Kubernetes sandbox environment for use with inspect_ai
An Inspect extension for agentic cyber evaluations
Accompanying code for Async Control: Stress-testing Asynchronous Control Measures for LLM Agents paper
Report Official Development Assistance (RODA)
Reproducing "Natural Emergent Misalignment from Reward Hacking" (MacDiarmid et al., Anthropic 2025) with open-source models. Includes reward-hackable RL environments, misalignment evaluations, training configs, and evaluation scripts. Models trained on OLMo (7B, 32B) and GPT-OSS (20B, 120B).