The company building AGI just told three safety researchers their concerns about AI risk were welcome—right after firing them for raising those concerns.
The Summary
- OpenAI fired three AI safety researchers—Jasmine Wang, Mikita Balesni, and Tomek Korbak—claiming they violated policies on handling sensitive information, not for voicing safety concerns
- The fired trio published an open letter warning their terminations would create a "chilling effect" on company culture around safety discussions
- OpenAI says there was a "significant breach of trust beyond what's outlined in the letter," but won't say what that breach was
- The timing raises questions about what "speaking out" means at a company racing toward artificial general intelligence
The Signal
OpenAI's response to the firings reads like a masterclass in corporate doublespeak. The company insists it "encourages those types of conversations and considers them essential to making the right decisions" about AI safety. Then it fires three people who had those conversations and calls it a policy violation.
The researchers didn't leave quietly. Their open letter, titled "OpenAI cannot make AI safe on its o" (the article cuts off mid-sentence, which is its own kind of metaphor), went directly to OpenAI's safety committees. They warned about a chilling effect. OpenAI responded by turning up the thermostat.
"Our internal investigation uncovered a significant breach of trust beyond what's outlined in the letter they published and we stand by the decision to not continue their employment."
Here's what we know: Three safety researchers raised concerns. They got fired. OpenAI claims it was for mishandling sensitive information. The researchers claim it was retaliation dressed up as policy enforcement. OpenAI insists the decision "was not about raising safety concerns or speaking out" but won't detail what the actual breach was.
This matters because of what OpenAI is building. Not a better chatbot. Not a smarter autocomplete. They're building systems they believe could reach artificial general intelligence. The safety researchers aren't there to make sure the product works. They're there to make sure it doesn't kill us all in the process.
The pattern is familiar:
- Company says safety is paramount
- Safety team raises concerns
- Company says concerns are actually policy violations
- Safety team gets walked out
Wang, Korbak, and Balesni aren't the first OpenAI safety researchers to leave under cloudy circumstances. The company has had a steady drip of departures from its safety and alignment teams. Each time, OpenAI says the departures are normal. Each time, the departed say something else.
The real signal here isn't the firing. It's the chilling effect the researchers warned about. If you're a safety researcher at OpenAI right now, you just watched three colleagues get fired for reasons the company won't fully explain. You know they raised concerns. You know the company says that's not why they were fired. You also know you can't prove what the company knows or doesn't know about the "breach of trust."
What do you do the next time you see something that worries you? Do you document it in Slack? Put it in an email? Raise it in a meeting? Or do you decide that whatever you're seeing probably isn't worth your job?
The Implication
OpenAI just taught every AI safety researcher in the building exactly what happens when you make the company uncomfortable. The official story is policy violations. The lived story is three empty desks and a message: be careful.
For anyone building in the agent economy, this is your reminder that the companies creating the foundation models your agents run on are optimizing for speed, not safety. The people whose job it was to slow them down keep getting fired. Plan accordingly.