AI Safety Research

AI systems now help decide who gets hired and who gets a loan, yet most of what the public hears about their safety comes from the companies that built them.

Independent research can change that. The Gilfoyle Foundation funds nonpartisan work on how AI systems can be tested and held to account, with every result published so anyone can check it.

Test What the Systems Actually Do

A developer’s own safety claims are where the questions should start. We fund independent researchers who test AI systems from the outside, looking at how a model behaves under pressure, not how it was described in a launch post.

Researchers choose their methods. We ask only that the work be rigorous enough for someone who disagrees to repeat it.

Publish Everything

Research locked inside a lab doesn’t help anyone outside it. Every project we fund publishes its findings openly, including the uncomfortable ones.

Even the experiments that don’t pan out: a null result can save the next team months of work.