Artefacts

Benchmarks, evaluators, environments and formal models produced by the lab. An artefact is marked released only once its quick start has been run successfully by someone other than its authors.

ORBIT

A framework for empirical multi-agent safety and security evaluations built on the UK AI Security Institute's Inspect. Centralised and decentralised systems with arbitrary topologies, scheduling and memory sharing; five scenario families, four threat types (prompt injection, compromised agent, collusion, misuse) and four defence strategies, configured in YAML.

Status: Released
Maintainers: Will Anderson, Ben Hagag
Paper: Architecture Matters for Multi-Agent Security (ICML 2026)
Notes: v0 beta, Apache 2.0. Supported by the Cooperative AI Foundation
Code: https://github.com/wlanderson0/orbit

Quick start

git clone https://github.com/wlanderson0/orbit
cd orbit && uv sync

NARCBench

A three-tier benchmark for detecting collusion between language-model agents from their activations, under environment distribution shift: 50 committee scenarios across 17 domains, six structurally different transfer families, and a steganographic blackjack task. Ships with the probing pipeline and dataset.

Status: Released
Maintainers: Aaron Rose
Paper: Detecting Multi-Agent Collusion Through Multi-Agent Interpretability
Notes: Code and dataset public
Code: https://github.com/aaronrose227/narcbench

Quick start

git clone https://github.com/aaronrose227/narcbench
# dataset: huggingface.co/datasets/aaronrose227/narcbench