AI control asks: if a model might be trying to sabotage us, can we still use it safely? ControlArena is the UK AISI’s open toolkit of settings, model organisms and protocols. Run one setting, then improve a monitor or document a gap.
Control is one of the most practical near-term safety agendas, and it rewards ordinary software and security skills.
Output: A documented experiment or a contribution to ControlArena.
Read the ControlArena README and run its example setting.
Write down the “red team” (attacker) and “blue team” (monitor) strategies in plain English.
Change one thing (a monitor prompt, a threshold, an attack) and measure the effect.
Open an issue or PR for anything confusing you hit along the way.
Write up your result in a short post.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Fix a small issue in a tool that safety researchers rely on: Inspect Evals, TransformerLens, ControlArena, nnsight or Neuronpedia. No ML research experience needed, just software skills.
Take a benchmark from a research paper (or another framework such as lm-eval-harness) and implement it in Inspect Evals, the largest open repository of AI safety benchmarks.
Pick a safety paper with public code and make it reproducible: fix broken dependencies, write setup docs, or package it so others can run it in one command.
