UKGovernmentBEIS/control-arena: ControlArena is a collection of settings, model organisms and protocols - for running control experiments.
ControlArena is a collection of settings, model organisms and protocols - for running control experiments. There was an error while loading. Please reload this page. This is a library for running AI Control experiments, created by the UK AI Security Institute. AI Control aims to design and evaluate control protocols. Control protocols are plans designed to prevent unsafe actions by AI systems, even if those AIs were misaligned and intended to subvert the safeguards. 💬 Community & Support: Join us in the control-arena-discussion channel on the Inspect AI Slack workspace. If you're not already on the Inspect workspace, please get in contact here and we can add you. This library includes: Install in your project: For most experiments you'll also want Docker, so that Inspect can create and manage sandboxes. See the Docker official installation guide for details. Finally, we recommend also installing the Inspect VSCode Extension so that you can easily run evaluations. In this example, we e
ControlArena ControlArena is a Python library for running AI Control experiments, created by the UK AI Security Institute and Redwood Research . AI Control aims to design and evaluate control protocols. Control protocols are plans designed to prevent unsafe actions by AI systems, even if those AIs were misaligned and intended to subvert the safeguards. 📚 Documentation : Visit our documentation site for detailed guides and API reference. 💬 Community & Support : Join us in the control-arena-discussion channel on the Inspect AI Slack workspace. If you're not already on the Inspect workspace, pl
related reading
- Settings – ControlArenacontrol-arena.aisi.org.uk
- [2604.15384] LinuxArena: A Control Setting for AI Agents in Live Production Software Environmentsarxiv.org
- The case for ensuring that powerful AIs are controlled — LessWronglesswrong.com
- GitHub - UKGovernmentBEIS/inspect_ai: Inspect: A framework for large language model evaluations · GitHubgithub.com
- Research Areas in AI Control (The Alignment Project by UK AISI) — AI Alignment Forumalignmentforum.org
- ARENA - AI Safety Curriculumlearn.arena.education
- Incident Report: unsanctioned agent behaviour during cyber testing | AISI Workaisi.gov.uk
- GitHub - microsoft/WindowsAgentArena: Windows Agent Arena (WAA) 🪟 is a scalable OS platform for testing and benchmarking of multi-modal AI agents.github.com
- AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
- Introduction to AI Control - by Sarah - BlueDot Impactblog.bluedot.org
- Research Areas in Benchmark Design and Evaluation (The Alignment Project by UK AISI) — AI Alignment Forumalignmentforum.org
- BashArena and Control Setting Designblog.redwoodresearch.org