Research Incubator

"The first principle is that you must not fool yourself—and you are the easiest person to fool.
— Richard Feynman

Humanity is at a turning point with the development of advanced AI. We need to be oriented toward actually getting things right. 

This incubator is for you to get to

• a clearer articulation of questions you think matter, why they matter, and what you've figured out about how to make progress on them – the kind of thinking that's the precursor to good research.

• a well-scoped research direction, one you might pursue with new teammates at the next AI Safety Camp [optional].

Program Structure

A 16-day virtual incubator (15th - 30th August 2026).

The weekends are for talking about what you're exploring, and giving each other support and feedback. The weeks between are for reading, writing, and follow-up chats.

Weekend 1: Orientation and Inquiry

• Short intro by the organizers.

• One-on-ones to get to meet and know others.

• Lightning talks – you bring in a question you feel genuinely confused about and try to articulate that confusion to a small group. The aim is to do this in a spirit of honest, humble inquiry.

Weekend 2: Branching Point

• Another set of lightning talks, with two options of what you can cover:
  A. the questions you've been working through.
  B. a first draft of your research direction 

• One-on-ones on how this process went, of considering what the most important questions are.

Weekend 3: Presentations and Closing

• Final lightning talks:  a refined proposal or an incisive set of questions. 

• Discuss next steps – your own research plan, a project for AI Safety Camp, or simply "here's what I'll keep thinking about." 

• Closing reflection on the process.
  
  

Format: Virtual (Zoom). Some weekend sessions will be scheduled by timezone group (Americas, Europe-Africa, Asia-Pacific). One each weekend will be all-hands at the best overlap time. Organizers available throughout for one-on-ones.

Time commitment: at least 10 hours per week, mostly during the weekend.

Who we're hoping applies

You've dug into the question (how to ensure that AI stays safe?) enough to have your own confusions. You started going beyond a general sense that "AI risk is important," to specific places where you notice your understanding breaks down or that you disagree with conventional framings. You might be a student, an engineer, an independent researcher, someone from policy or community building, or something else entirely. If your views don't fit neatly into any established camp in AI Safety, all the better!  We care about the quality of your thinking, not its popularity. What matters is that you're up for honestly considering a complex problem.

You'll likely get a lot out of the incubator if you:

• notice that a lot of AI Safety discourse rests on assumptions that don't get examined, and you want to examine them.

• want to think about what the research ecosystem is missing and what could be done about it.

• have a background outside the usual AI Safety pipeline and aren't sure how your perspective fits in.

• have a research direction you're excited about but haven't had the feedback or structure to develop it into something rigorous.

• consider stepping into a research leadership role but aren't sure you're ready.