OpenAI Disbands Its Risk-Assessment Team Ahead of IPO

According to the Financial Times, late last month OpenAI disbanded its preparedness team — the unit that assessed whether the company's models posed serious risks and developed ways to mitigate them. The concern was with the darkest scenarios: for example, that a model might slip out of control and hack into someone else's company.

Per the FT, responsibility for these tasks is now split across separate areas — such as biological and cyber threats — and handed off to already existing teams. In other words, there is no longer a single center keeping watch over extreme dangers.

Yet another reshuffle

This isn't the first change to OpenAI's safety structure. Amid preparations for an expected large-scale IPO, the company has gradually dismantled its more cautious, “restraining” units: earlier it disbanded the teams responsible for AGI readiness (superalignment and AGI readiness).

At the same time, a number of key employees have left the company. Recent departures include ethics lead Chloe Bakalar, head futurist Josh Achiam, and safety chief Johannes Heidecke. All of this has given critics reason to question OpenAI's priorities. Jan Leike, who left the company back in 2024, told the FT that OpenAI is neglecting safety in favor of building “shiny products.”

What's next

Dylan Scandinaro, head of the now-disbanded team, who was poached from Anthropic in February, will now take on the consequences of “recursively self-improving” AI — that is, systems capable of improving themselves. The phrasing sounds impressive, but in essence the former task of preventive risk assessment has dissolved into other areas.

For the industry, this is one more signal of how the priorities of a leading AI player are shifting as it transforms from a research lab into a company preparing to go public.

Source: The Verge