Geoff Ralston, former president of Y Combinator and one of the startup world’s most respected operators, is back in the game — and this time, he’s betting on AI safety.
Meet SAIF — the Safe Artificial Intelligence Fund.
The acronym isn’t just branding. SAIF’s thesis is clear: invest in startups focused on the safe, secure, and responsible deployment of AI technologies. And yes, the fund will write $100K SAFEs (simple agreements for future equity) — pun fully intended — capped at $10 million valuations.
“Most AI startups today are solving problems, building capabilities — but safety isn’t always the priority,” Ralston told TechCrunch.
“I want to back the ones who put safety at the core.”
What SAIF is backing:
Ralston is casting a wide net, with a few clear red lines. He’s looking at startups working on:
AI explainability and benchmarking
Compliance & audit tools for AI
IP protection and secure AI architectures
AI disinformation detection and cyberattack defenses
Tools that make AI decisions safer in business environments
Even “AI weapon safety systems” — designed to prevent misuse or unauthorized deployment — are in scope.
What he won’t touch?
Fully autonomous weapons or AI designed for offensive warfare.
“AI managing weapons without a human in the loop? That’s where I draw the line,” said Ralston.
Why it matters:
As VCs flood the AI space, SAIF stands out by focusing not on speed or scale, but trust and security. In an ecosystem sprinting toward the next big thing, Ralston is taking a principled pause — funding the infrastructure that could prevent the future from going off the rails.
The YC edge:
Ralston isn’t just cutting checks. He’s offering mentorship, YC app coaching, and access to his vast investor network — a serious value-add, especially for early-stage founders who want the YC halo without getting lost in the crowd.
He hasn’t revealed how large the fund is, how many bets he’ll make, or who the LPs are — but if Ralston’s track record is any indication, expect some sharp founders to come knocking.
In a moment where AI can be anything, Ralston’s betting on AI that won’t become everything, everywhere, all at once — without oversight.