The Safety Funding Gap
Money is a map of priorities. Capabilities get the capital. Safety gets the change jar.
$250B+
Global yearly investment in making AI more powerful (industry scale, mid-2020s estimates).
~$190M
Yearly flow into technical safety, evaluations, policy, and advocacy combined (compiled public estimates).
More than 1,000×
Less than one-tenth of one percent of the money building this technology goes toward making sure it does not end us. Inside the safety pot, advocacy and public engagement are a thin slice on top of research and evals.
Where safety money goes
Rough split: technical research largest; evaluations next; governance and policy smaller; field-building and advocacy smallest. The political work that produces treaties is underfunded relative to the stakes.
What to fund instead of theater
Independent evaluations outside lab walls. Treaty design and diplomatic capacity. Public advocacy that creates permission for hard law. Compute-governance tooling. Lab-internal safety teams alone cannot write international rules.
Figures are rounded midpoints compiled from public sources (AI safety funding overviews, 80,000 Hours, Open Philanthropy grant analyses, Inside Philanthropy). Published totals vary with definitions. See nakadafoundation.org/blog/why-is-there-so-little-ai-safety-funding/ for the working.