If you expect a large liquidity event in the next year or two — an AI company IPO, a tender offer, a secondary — and you intend to give some of it to AI safety, you should borrow against that money and give it away now. As far as I can tell, almost nobody has done this. That is strange, because borrowing against future liquidity is one of the most ordinary financial acts there is. People borrow to buy houses they cannot yet afford, to get college degrees that won’t pay off for a decade, to buy cars, to start companies, to buy out a partner. This is not reckless, this is normal, boring financial behavior.
The mechanics are completely straightforward. If you already hold public securities, you can use a pledged asset line or securities-backed line of credit at your broker: setup takes days, and all-in rates currently run in the 6.5–9% range depending on the lender and the size and diversification of the collateral. If your equity is still private, several firms write non-recourse financing against pre-IPO shares and options, repaid out of exit proceeds, with your other assets ring-fenced if the exit never comes. Do note that many companies’ insider trading and transfer policies prohibit pledging shares, so check before you plan around it. And non-recourse debt that gets forgiven can generate a tax bill on the forgiven amount. Talk to your financial advisor.
Borrowing against your future liquidity will allow you to spend more on safety now. Money spent on safety today is worth considerably more than the same money spent in two years. If you believe slowing AI down would be good, the direct corollary is that speeding up the countervailing work is also good, and money now buys more of that per dollar than money later. But more importantly, if you expect to spend a lot in the future then you better be spending now. You have to spend money to spend money. People keep asking me how to deploy $100 billion in a year or two. The answer is one dollar at a time, starting now. If you want to fund large-scale projects later, you have to seed them now. If you want excellent people running them, you have to find those people and cultivate them now. If you want a breakthrough funded, someone has to have identified and de-risked the problem first. This is how capital works. When the Future Fund collapsed, many expected Open Philanthropy to spend less, because $100M+ had already flowed into the space. Instead they spent more because that money had created fundable opportunities that did not previously exist. Funding is a flywheel: you fund things, funded things generate more things worth funding, and the whole thing scales.
You can also deliberately optimise for this. You can fund talent development. You can fund a project past the point of comfort specifically to find where its absorptive ceiling is, and then scale back. You can make career transition grants: ask everyone in the field to name the most talented person they know outside it, and fund that person to move and roughly double the community overnight. You can identify the parameters you are most uncertain about and buy experiments that test them, rigorously or informally. You can fund a bouquet of bets and cull the ones that fail. The idea that you work out what to fund by sitting in an armchair and thinking hard is incorrect. You work it out by funding things, watching what happens, and iterating. Philanthropic funding is more like software engineering than it is like philosophy.
Here are some real gaps: I put out an RFP on power concentration that with applicants god enough to deploy $20M at a very high bar; three dinners in, we have raised $5M, and the next $15m in opportunities are all shovel ready. Last year the UK AI Security Institute’s Alignment Project could have absorbed another $50M and nobody stepped up. Effective Institutions Project has an exceptional track record in AI safety (and in democracy) and could regrant tens of millions over 6 months; so could a number of individuals I would trust with it. I am personally sitting on around twenty grants that look worth funding at some level and cannot fund them. In my personal capacity I advise a few donors on political giving and I come across tens of millions of dollars of 501(c)(4) and political opportunities a year. Juniper Ventures aims to raise a billion-dollar venture fund to keep for-profit safety companies well-funded with aligned capital — Goodfire, Apollo, and others — and there will likely be room in it. Halcyon Futures could absorb more for talent and for its biosecurity work. One of the key people working on verification told me recently he could deploy $600M this year and next. The bottleneck is not a shortage of things to fund. It is that a very small number of evaluators are charged with assessing everything, so the money moves too slowly.
We should listen to the outside view here: Nobody spends money as fast as they intend to. The Giving Pledge turned fifteen last year, and of the fifty-seven original signers exactly one living couple has fulfilled it; the same research found that nearly three-quarters of billionaires have given away less than 5% of their wealth. The Future Fund had an entire ecosystem working for it at maximum ambition and moved only $100M in nearly a year before it collapsed. The Gates Foundation has done the best out of any foundation in history, but managed to spend $100b only by scaling for twenty-five years to a staff of 2,000. Deployment is hard. You should expect to underspend relative to your own stated intentions, by a lot. No philanthropist in history has ever spent faster than they wanted to, so you should try to spend faster than feels comfortable.
Borrowing to give might seem intuitively reckless to the risk averse person. But borrowing to give reduces philanthropic risk — the risk of accidentally funding bad things. If you wait, you will predictably end up in a situation one year from now where you have tons of money, AI going off the rails, and very little good work to fund. Predictably, you will want to throw money at random bets and take crazy Hail Mary’s to try to fix things. If you don’t do that, some other funder will. Funding now produces evidence about what works, and it crowds the bad projects by creating room to scale good ones. Going from $2B one year to $100B the next produces an insane level of distortion which can be avoided by ramping. But ramping to this number requires philanthropic scaling ambition the likes of which we have never previously seen!
The argument is even stronger if you believe, along with the median member of technical staff, that your stock is underpriced but want to help AI safety. Then you make returns by borrowing at market rates instead of selling any stock.
If you want introductions to any of this — the power concentration RFP, EIP, political giving, Juniper Ventures, Halcyon Futures — email me at tyler@foundation-layer.ai and I will make them happen.
Runtime Verification 
Leave a comment