Thoughts on where to Donate for AI Safety in 2025

All views here are my personal views. They do not represent that of the Long-Term Future Fund nor any other past or future employers. Written quickly. I intend to crosspost on the EA Forum and other venues after significant feedback.

I’ve been working as a grantmaker for the Long-Term Future Fund since the beginning of 2022, and have been directly responsible for evaluating several million dollars of grants, and significantly influenced some more. In a professional or semi-professional capacity, I’ve also spent a lot of time communicating about grants: I’ve written a number of posts on AI Safety, longtermism, and charitable donations for the EA Forum, including the most popular post on the “grantmaking” topic.

Most of my learnings are either very abstract and “meta” or specific and technical to specific situations that might come up, but occasionally people (very reasonably) ask me a specific question:

• Where’s the best place I can donate to?

In general, historically my answer has come in 3 broad categories:

1. Eh, I don’t know man..

2. Donate to the Long-Term Future Fund, maybe?

3. [less common] Specific place I/LTFF evaluated that seems like a good donation opportunity at the time

Today, I have different answers! Not necessarily useful, but different:

1. Politics, maybe? (for Americans and American green card holders)

2. LTFF is still a decent shot

3. Investing your money, as long as you can reasonably commit to donating it in the future

For Americans and American green card holders thinking of donating between $5,000 and $100,000 (possibly more), I think I mostly buy the arguments presented here. The lower bound is considering the impacts of non-financial costs of political donations, the upper bound is thinking of contribution limits.

I think it’s very plausible that people should aim their donations to cause better AI Safety policies to happen. In particular, donations to Alex Bores’ and Scott Wiener’s campaigns before the end of 2025 seem unusually high-leverage right now.

I think the cost-effectiveness of donating to their campaigns seem significantly higher than either other political campaigns or technical AI safety or AI policy research right now, with moderate confidence. I think Eric’s arguments here are pretty decent (cf: Michael Dickens’ epistemic spot-checks).

Please all note that political donations are generally not tax-deductible, however tax-deductibility should not affect where you donate to all that much (basically if your marginal tax rate is <50%, tax-deductibility shouldn’t affect the cost-effectiveness of your donation choices by more than a factor of 2).

Some other caveats: I’ve investigated this in less detail than I would like, political donations are in some ways more messy and have more career considerations than purer charitable donations, and there’s a high probability your donations won’t achieve anything concrete.

I think the long-term future fund is still a decent vehicle for donations. We still receive good applications for individuals and projects that would be great to seed or continue to fund. In many ways we are still the best source and recipient of funding for small projects, though Manifund and Longview are also decent options.

For idiosyncratic reasons, LTFF fund managers are less able to spend time on investigating our grants. I think this decreases the effectiveness of our grants, but not hugely (~5-25%). There may also be structural changes to LTFF in 2025. I’ll write more when I can. I think this affects the long-term trajectory of donations but not necessarily the cost-effectiveness of donations to us.

Ultimately, for the direct cost-effectiveness of the marginal grant, I think donations to LTFF are likely to be slightly higher this year than the past. This is because we aren’t fundraising as much, and thus expect to receive far lower donations relative to applications. However, I personally believe this effect is dominated by the potentially significantly higher cost-effectiveness of political donations, for the marginal donation.

LTFF fund managers also do active grantmaking under the AI Risk Mitigation Fund, which has less of an established history but makes similarly effective donations (and are entirely focused on AI).

I think there’s a strong case to be made for waiting and investing now, and re-evaluating the landscape in 6-12 months. I think the case is strongest for re-evaluating political donations later, but plausibly we’ll have better technical AI Safety and policy options to give to later as well.

I think now is an unusually confusing time, at least for me. Perhaps we will all be less confused over the course of the next year.

There’s the major caveat that it’s also very plausible that many other people have similar considerations, in addition to a bunch of AI companies vesting, and there will be an inflow of money to AI safety and adjacent causes in the coming year.