AI Safety Is Hiring — Your Management Skills May Beat a PhD
The field has money and big ideas, but almost nobody to run things.
The AI safety field isn't constrained by money, talent, or ideas — it's short of competent, agentic people who can get good work done with high independence. That's the argument from someone who spent the last two years as a quant trader, did MATS (an AI safety research programme) this summer, and then decided to work on reducing catastrophic AI risks. He estimates roughly 400 people a year attempt to build new safety orgs, against 1,000–5,000 working on AI risk in total. And exploration is cheap: about 30 hours to get properly informed about the risks, maybe 60 hours to do substantial further derisking like a small project, and 6–12 months to test working in the field more seriously.
- The bottleneck isn't scientists — it's organizers, founders, and managers who finish what they start
- Roughly 400 people a year try to build new AI safety groups, out of only 1,000–5,000 working on AI risk total
- You can test the water cheaply: about 30 hours of reading, or 6–12 months of part-time work, before quitting anything
Why It Matters
If AI risks are real, the people deciding how to handle them are being picked in the next year or two.