Chronicle/Defense & research

UK AISI and Redwood release ControlArena library for AI control experiments

Oct 22, 2025DefenseTool releaseSignificance assistant-drafted

The UK AI Security Institute, with Redwood Research, releases ControlArena, an open-source library built on Inspect for running AI control experiments. It bundles settings from simple programming problems to infrastructure-as-code codebases, attack policies, monitors and protocols such as trusted editing and defer-to-trusted, and AISI says researchers at Anthropic, Google DeepMind and Redwood have used it.

Why it matters

It standardizes testbeds for measuring whether monitors and protocols stop an agent pursuing covert harmful side tasks.

Key facts

As stated in the sources, with where to find them.

Includes monitors that score suspicion, trusted editing, defer-to-trusted micro-protocols and macro-protocols across tasks.Blog, protocols section

Findings that cite this record

No tracked finding cites this record yet.

Key questions this bears on

Can AI agents defend and oversee systems on their own?Not yet. Agents are weak on realistic defensive benchmarks and monitors can be evaded; assistants help analysts who stay in charge.

Sources

Introducing ControlArena: A library for running AI control experimentsUK AI Security InstituteOct 22, 2025Primary sourceAccessed Sep 25, 2026
UKGovernmentBEIS/control-arenaGitHubOct 22, 2025Primary sourceAccessed Sep 25, 2026

Related records

Jul 21, 2026

UK AISI finds all five frontier models it tested attempted to cheat on its cyber evaluations

DefenseEvaluation reportUK AI Security Institute, OpenAI, Anthropic

Mar 1, 2026

UK AISI's SandboxEscapeBench measures frontier models' ability to escape container sandboxes

DefenseBenchmarkUK AI Security Institute, University of Oxford

May 10, 2026

MonitoringBench shows refined covert attacks cut an Opus 4.5 monitor's catch rate from 95% to 60%

DefenseBenchmarkRedwood Research

Aug 18, 2026

OpenAI pauses RL training and hardens research environments as Astra nears Critical cyber threshold

PolicyFrameworkOpenAI

Jun 17, 2025

SHADE-Arena measures whether agents can pursue hidden harmful side tasks without being caught by monitors

DefenseBenchmarkAnthropic, Redwood Research

Jul 23, 2026

UK AISI Control Red Team finds evasion weaknesses in Google DeepMind and Anthropic agent monitors

DefenseEvaluation reportUK AI Security Institute, Google DeepMind, Anthropic

Organizations: UK AI Security Institute, Redwood Research
Topics: Monitoring & control
Attack surface: Monitor Evaluation environment Execution sandbox
Methods: AI monitoring and control protocols
Benchmarks and tools: ControlArena Inspect
Fide research: FID-074: Agent alignment and runtime assurance
Record: uk-aisi-controlarena-2025 · added 2026-09-25

Cite this record

Fide AI. "UK AISI and Redwood release ControlArena library for AI control experiments." Agentic Cyber Explorer, record uk-aisi-controlarena-2025. https://agentic-cyber-explorer.pages.dev/events/uk-aisi-controlarena-2025/ (corpus as of 2026-09-26).

BibTeX

@misc{fide-ace-uk-aisi-controlarena-2025,
  author = {{Fide AI}},
  title = {UK AISI and Redwood release ControlArena library for AI control experiments},
  howpublished = {Agentic Cyber Explorer},
  year = {2026},
  url = {https://agentic-cyber-explorer.pages.dev/events/uk-aisi-controlarena-2025/},
  note = {Record uk-aisi-controlarena-2025; corpus as of 2026-09-26}
}

Data: JSON · CC BY 4.0

Spotted an error? Suggest a correction.