Accelerating Safely
The future will be dominated by AI, and that future is rapidly approaching. Whether or not to accelerate into that future is no longer the question. So here’s the challenge: Can we accelerate safely?
SAIF, the Safe AI Fund (saif.vc) is a small, focused effort to help exceptional founders build products that support a future where AI is both powerful and aligned with human interests. We believe that AI will soon permeate every part of society—our economies, our institutions, and our daily lives. The stakes have never been higher.
We’ll invest early, help shape go-to-market, and use our experience to increase the odds of downstream success. I will be hands-on with each startup, including weekly office hours in order to help them prioritize, find product market fit, raise seed capital, and figure out how to make something their customers want.
We will fund a diverse group of companies; if it is related to AI safety and / or security, we’re interested. For example, we’d love to see companies which will:
Dramatically improve information security.
Improve consumers’ ability to identify disinformation.
Build tools which provide advanced models with improved alignment and/or interpretability.
Build superhuman forecasting tools.
Build tools to provide advanced warning of biothreats or new pathogens.
Build better negotiation tools for trade, treaties, and more.
It’s a broad space, and we’re open on domain - but very selective on people and ideas.
There is a lot of disagreement over our future with AI, although we can surely all agree we want to get there safely. Yet somehow, “safe” has become a loaded term in today’s AI discourse. That’s unfortunate because safe and secure software shouldn’t be controversial—it should be foundational.
The enormous potential of AI is unquestionable, but the possibility remains that AI technology could lead to devastating, even ultimate harm. The most fervent techno-optimists and most fearful pessimists agree on this basic point.
SAIF is not a reaction to hype or fear. It’s a bet on a future that is both accelerated and aligned—and on the founders brave enough to build it.
This hits close to home. Over the past year I've had hundreds of conversations with engineering and platform teams at Google Next, AWS, KubeCon. The pattern is always the same: teams want to deploy AI but need guardrails that reflect their own policies, not generic safety filters. Every organization has different rules, different compliance requirements, different contractual obligations, and no way to enforce them across what their AI actually produces.
Most of the safety conversation focuses on the model layer. But the gap I keep seeing is at the organizational layer, where companies already have the rules written down in documents nobody reads, and no system connects those rules to the AI doing the work. That feels like the most underbuilt piece of the "accelerate safely" stack right now.