AISC Research Incubator

Raising $5,000active

A 16-day virtual incubator (Aug 15-30, 2026) for developing sharper research epistemics in AI Safety and arriving at well-scoped project ideas. 20-30 participants.

Project Details

A lot of discourse in AI Safety rests on assumptions that are seldom examined. In conversation, that costs little. However, once researchers set out to collaborate, it becomes essential to both engage with the full difficulty of addressing AI risk and find a project that is pragmatically tractable. It’s not enough to rush safety work in the wake of LLM progress. Given the magnitude of the risk, we need to be oriented toward actually getting things right.

This program is designed to give participants the structure and feedback to do that work seriously.

We will run a 16-day virtual incubator (15-30 Aug 2026), which will be hosted via Zoom and Slack, organized by the two AISC organizers Robert Kralisch and Remmelt Ellen. The incubator is structured around three weekends, with self-directed work during the intervening weeks. The intended participant number is between 20 and 30, and applications are already at 58 (we expect at least a third more around the application deadline).

On the first weekend, participants receive an orientation from the organizers on what we consider to be markers of high-quality AI Safety research – clarity of assumptions, well-articulated theory of change, realistic scoping, and honest engagement with uncertainty. Participants then give lightning talks in small groups, each presenting a question they are genuinely confused about in AI Safety and attempting to articulate that confusion clearly. The aim is to establish a culture of honest inquiry from the outset.

During the first week, participants read, write, and engage in follow-up conversations – developing the ideas from their lightning talks and any threads that have come out of it. We’ll encourage people to connect with others who share an interest in similar questions, and support them in developing their understanding. Every participant has at least one 1-on-1 call with an organizer during this period, focused on helping them identify where their thinking is sharpest and where it needs more work.

On the second weekend, participants give a second round of lightning talks, choosing between two tracks: (A) a deeper articulation of the questions they've been working through, or (B) a first draft of a research direction. We treat "I now have better questions" and "I now have a concrete proposal" as equally valid outcomes at this stage. This is the most flexible stage of the program, as we intend to use the second weekend to address any patterns we noticed among participants, providing further orientation, recommended reading, and guidance on forming collaboration or finding discussion partners.
We also plan to invite 1-2 AISC alumni who led strong projects as guest speakers, to give participants concrete examples of what a well-executed project trajectory looks like.

Organizer 1-on-1s continue in the second week, now focused on either sharpening the question-set or scoping the research direction. We intend for most participants to either join in discussion groups or demonstrate productive autonomy.

In the third weekend, participants present their final output – either a refined research proposal or a well-articulated set of research questions – and discuss next steps. These may include pursuing the research independently, applying to lead a project at the upcoming AI Safety Camp, forming a team with other incubator participants, or continuing to develop their thinking informally.
We are planning on encouraging participants to publish their outputs via posts or the AISC website, depending on how productive we expect this to be for them.
We will also run a survey at the end of the program to get participant feedback on how helpful they found it, as well as on potential improvements to the format.

The primary measure of the incubator's success is whether participants go on to submit stronger project proposals in the upcoming AISC project lead application round, though we also consider it a success when a participant leaves with sharper questions and better research orientation.
We intend to do a lightweight survey 6 months after the program with participants who have not applied to lead a project at AISC, to ask whether the incubator had an impact on their career trajectory.

We expect at least 5 project lead applications to emerge directly from incubator participants, compared to a current baseline of about 45 proposals per round. We'll also track whether incubator participants' proposals require fewer revision cycles before acceptance.

Theory of Impact

The purpose of this program is to help address two bottlenecks that AI Safety Camp faces:

We notice that many project-lead applicants have confused assumptions about AI risk, and struggle to articulate a high quality theory of change for their project. This is not typically their fault: the field just hasn’t established good norms for what a well-grounded theory of change in AI Safety research looks like, and this shows up downstream.
While AISC is happy to host project work that is “merely potentially useful”, especially if it can serve as an important learning experience that accelerates participants’ transition into AI Safety work, we find it important that project leads have clarity about how their work is situated in the overall effort to avoid AI-driven catastrophic risk – not least because this affects the same calibration in their team members.
AI Safety Camp is short on high quality project proposals, relative to the number and overall qualifications of team member applicants.
We have consistently received feedback from project leads about how they were impressed with the qualifications and competence of team members, and we think this applicant pool is an important resource we’d like to make full use of. Increasing the number of high quality project proposals would increase the competitiveness of that stage of the application process in a healthy way, and allow us to scale in good conscience to the interest level already demonstrated in the amount and quality of the team member applications.

People

Robert Kralisch

Team Member

Grants Received– no grants recorded

Add grant

Funding Asks

grantmaking.ai Launch Round

Applied

Minimum

$5,000

Ideal

$30,200

What this ask is

How the money will be spent

Mostly paying for organizer work:
Robert Kralisch $10,000

Remmelt Ellen $10,000

Experienced inquiry facilitator $10,000

Software $200

Discussion

No comments yet. Be the first to share your thoughts.