Take action

The outcome isn't fixed. People shape it.

AGI risk is unusual among global problems: it is neglected, tractable, and talent-constrained. A relatively small field is working on it, much of the most useful work is bottlenecked on capable people, and progress is real. Here's how to contribute — without making things worse.

First, do no harm The most common ways to be counterproductive: accelerating raw capabilities under a "safety" banner, and spreading low-quality alarmism that discredits the field. The highest-value contribution is usually deep expertise in a tractable sub-problem, applied where the field is bottlenecked — not volume.
If you're technical

Do the research

Alignment, interpretability, evaluations, and AI control are talent-constrained. Adjacent skills — security, ML engineering, distributed systems, formal methods — transfer directly. Entry routes:

If you're policy-minded

Work on governance

Governance is high-leverage and needs lawyers, economists, communicators, analysts, and civil servants — not just technologists. Areas that need people: compute governance, evaluations and standards, liability, and international coordination.

  • GovAI, CSET, IAPS — research & fellowships
  • National AI institutes (UK AISI, US CAISI)
  • Legislative and regulatory roles
If you want direction

Get advice & upskill

Free, high-quality guidance exists for finding your highest-impact contribution.

If you have capital

Fund the work

Safety and governance work is funding-constrained relative to capability investment. Established grantmakers route funds to high-impact projects:

If you want to advocate

Raise the quality of the conversation

Public understanding and political will shape what governance is possible. Useful advocacy is accurate and specific — not hype.

Whoever you are

Stay informed & share well

The single most accessible contribution: understand the arguments well enough to discuss them accurately, and point others to credible sources rather than hype.

Get in touch

Questions, corrections, or contributions

This is a living reference. If you've spotted an error, have a source to suggest, or want to collaborate, send a note. Accuracy matters here — corrections are genuinely welcomed.

We aim to keep every claim sourced and current. Flagged issues are reviewed and the page updated.