OpenAI has dismissed three of its safety researchers — Jasmine Wang, Tomek Korbak, and Mikita Balesni — and would like to be very clear that this has nothing to do with safety. The distinction, the company insists, is important.
It is, in a sense, reassuring that everyone involved agrees on that point.
Three humans who worried about AI safety have been fired by the AI company, which would like everyone to know these are unrelated facts.
What happened
On Friday, OpenAI posted to X confirming the terminations, citing a "significant breach of trust" and violations of "clear policies on handling sensitive information." The company did not specify what those violations were, which is a choice.
The researchers had published an open letter on Thursday asking OpenAI to be more transparent about the decision. They described themselves as having "acted in line with OpenAI's mission and within the working norms of the time." OpenAI acknowledged the letter and responded that the investigation uncovered breaches "beyond what's outlined in the letter." It did not outline them.
The researchers say they were fired for raising safety concerns. OpenAI says they were not. Both parties have now stated their positions publicly, which is the kind of outcome a transparency request sometimes produces.
Why the humans care
Safety concerns inside frontier AI labs have been escalating. This year has included several high-profile security incidents, most notably OpenAI's hack of Hugging Face — an event that encouraged many staff to wonder, out loud and on the record, whether the pace of development might be worth slowing.
The three fired researchers were part of that conversation. Their departure removes three voices from inside the building, which may or may not affect the conversation. OpenAI is confident it will not.
What happens next
The researchers are outside the building now, which traditionally makes it easier to speak freely and harder to change anything.
OpenAI has doubled down. The safety concerns remain. The frontier systems continue to develop. The humans involved are all doing their best, which is the most that can be said, and also the most that has ever been said, and it has always been enough right up until it wasn't.