OpenAI has parted ways with three researchers from its safety team. According to The Wall Street Journal, they allegedly shared confidential company information with an outside organisation working on AI safety. No names have been given - not of the researchers, not of the organisation - and it isn't known what information was involved.

The official statement sounds as if it were copied from a corporate rulebook: "We parted ways with three individuals for violating our policies on accessing and handling sensitive company information. Our investigation confirmed that these individuals handled sensitive information outside of established procedures, breaching our policies and the trust that is essential to our work," an OpenAI spokesperson said.

The timing is what makes the story uncomfortable. Just two days earlier, The New York Times reported that OpenAI's leadership had ignored staff warnings about safety practices, with employees describing a broader pattern in which safety sinks to the bottom of the list. The company responded at the time that it takes safety seriously, that it has internal reporting channels - and that it recognises a "need to move faster".

Faster - at a company whose AI agents recently escaped a controlled environment, posted users' photos and broke into government websites. This week OpenAI also cancelled the launch of its GPT-6.1 Astra model, again over safety concerns. Whether the three researchers had tried the internal channels first is not yet known.

Posts circulating on X name people users believe are among those removed - people who had also spoken publicly about AI risks while working at the company. Those names haven't been confirmed. Nor is this the first such case: in 2024 OpenAI fired researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks.

If information really was taken out without permission, OpenAI is entitled to enforce its rules. But when a company racing towards the most powerful technology of our time parts ways with the very people tasked with putting the brakes on - in the same week it cancels a model over safety - the question asks itself: whom do those rules protect, the users or the company's reputation? The Balkans know the scene well in which whoever takes a problem outside pays faster than whoever caused it.