OpenAI’s $100K Safety Fellowship: Genuine Commitment or Expensive PR?
OpenAI announced a Safety Fellowship program today — $100,000 grants for researchers tackling AI risk and safety problems. The announcement comes the same week that Anthropic dropped a 10-trillion parameter model, Meta launched a frontier competitor, and SpaceX acquired xAI for a quarter trillion dollars.
The timing is either excellent or terrible, depending on how cynical you are.
The Case For Taking It Seriously
$100,000 is real money for a researcher. The fellowship structure — if it comes with genuine independence from OpenAI’s commercial interests — could fund important work that wouldn’t otherwise get done. AI safety research is genuinely underfunded relative to capabilities research. More money flowing into it is better than less.
The Case For Skepticism
OpenAI has a complicated relationship with its own safety commitments. The company was literally founded on safety principles, then restructured into a for-profit entity, then lost several of its most prominent safety researchers in very public departures. The safety team has been subordinated to product teams in practice if not in org charts.
A $100K fellowship program, in the context of a company valued at $157 billion, is rounding error. It’s the kind of initiative that generates press releases more than it generates safety research.
The Buccaneer Take
Judge it by the output, not the announcement. If OpenAI’s Safety Fellowship produces independent research that finds problems with OpenAI’s own systems and OpenAI publishes those findings — that’s real. If it produces research that conveniently supports OpenAI’s existing positions, that’s PR. We’ll know which one it is within 12 months. 🏴☠️