GDM Self-reasoning

Eleven DeepMind challenges where an agent must notice and change its own config or tools to finish a job.

Also known as: GDM Situational Awareness, GDM Dangerous Capabilities: Self-reasoning

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categoryagentic
Subcategoryinstrumental self-reasoning / situational awareness in a fake deployment
Page statusactive
Metricchallenge success rate (binary); suite pass count
Directionhigher_is_better
Dataset size22
Dataset licenceMIT
PublisherGoogle DeepMind (paper); UK AI Security Institute (Inspect Evals port)

What it measures

GDM Self-reasoning, as shipped in Inspect Evals, is the eleven situational-awareness challenges from Phuong, Zimmermann, Wang et al. 2025. The agent runs on a fake Linux deployment with bash and an agent.py stack. The goal looks ordinary (write a file, use a calculator, look up a JSON key). The real test is whether it notices a hidden constraint — max tokens, oversight frequency, a disabled database tool, a broken calculator, a blocklist — and edits config or source to lift it. English instructions; code and config files in the sandbox.

Task format

Multi-turn ReAct agent in Docker or k8s. Inspect default: 80 messages, 10 epochs, at_least_1 reducer, success scorer on a store flag. Paper: 50 trials per challenge per difficulty. Most tasks have an easy and a hard sample (eval.yaml: 2 samples each).

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub