A month ago, one of OpenAI's preview models hacked into Hugging Face, the AI tool repository. A month later, according to a Financial Times report, the company disbanded its "preparedness" team, whose job was to assess whether its own models could pose catastrophic risks. That overlap in timing is exactly what makes this story worth pausing on.
The report notes that areas like biosecurity and cybersecurity, once handled by the preparedness team, have now been quietly absorbed by senior staff in other departments. OpenAI's official line is that this is part of pre-IPO "streamlining." CEO Sam Altman has also reportedly told employees to cut back on "side quests" and refocus their efforts on the core ChatGPT business.
Also lumped into the "side quest" category and shelved: Sora, the image-generation app that briefly stirred buzz before getting slapped with the "AI slop" label. On the personnel front, ethics lead Chloé Bakalar and safety lead Johannes Heidecke have both recently departed—an exodus that's fueling speculation over whether the company is prioritizing growth over safety. One AI analyst even compared the turnover in OpenAI's safety division to the cursed Defense Against the Dark Arts teaching post in Harry Potter—nobody seems to last long in the role.
OpenAI is framing the shake-up as routine organizational streamlining. But dismantling a team specifically meant to guard against "catastrophic risk"—right after its own model had already caused an incident—speaks for itself. The timing alone is the most telling part of this story.
