What happened
Three safety researchers fired by OpenAI last week have published an open letter. Jasmine Wang, Tomek Korbak, and Mikita Balesni deny the company’s claims that they mishandled sensitive information.
According to a report from TechCrunch, OpenAI stated the dismissals followed an investigation into a “pattern of misconduct”. The researchers argue their actions were previously part of OpenAI's culture and essential for safety work. They warn their firing creates a chilling effect on internal dissent.
How the room's reading it
This is a classic 'he said, she said' situation, but with high stakes for AGI safety. On X, safety-conscious developers are siding with the researchers, viewing the firings as a sign that OpenAI is prioritising speed over caution. Jasmine Wang's thread explaining her side has been widely shared.
The consensus among AI policy watchers is that this incident weakens trust in corporate self-governance. It provides fresh ammunition for regulators who argue that frontier labs can't be left to police themselves — especially after recent safety incidents at the company. Some lab insiders, however, are quietly suggesting that clear rules on information handling are necessary and that the researchers may have crossed a line, even if their intentions were good.
Sailfish's take
We've seen this movie before. When a research lab starts acting like a fortress, it's usually a sign that shipping product has eclipsed everything else. The specific claims of misconduct feel less important than the cultural signal OpenAI is sending — don't rock the boat.
This chilling effect is real for builders. If a lab’s own safety team feels muzzled, how can external developers trust the platform's guardrails? It pushes the burden of safety downstream onto people building applications, who have far less visibility into the model's core. We'd be wary of building critical systems on a platform where the internal safety culture is so publicly in question. The useful question isn't who is right, but whether you can build with confidence when the foundations are shaking.