- Published on
OpenAI reportedly disbanded its preparedness team
- Authors
- Name
- DeepHorus
OpenAI has reportedly disbanded its preparedness team, according to the Financial Times, at the end of last month. The team's mission was to assess whether models posed serious risks—like going rogue and hacking another company—and to develop mitigation strategies. This move marks the latest shakeup of the company's safety teams as it races toward a massive IPO. Responsibility has been divided across specific areas like bio and cyber threats, then folded into existing operational teams.
The disbandment follows a broader trend within OpenAI, which has gradually dismantled its more research-led safety structures. Over the past few years, it dissolved its AGI readiness and superalignment teams, signaling a shift away from centralized safety oversight. The preparedness team itself was relatively new, led by Dylan Scandinaro, who was poached from Anthropic in February. According to the FT, Scandinaro will now focus on the implications of 'recursive self-improving' AI, a niche but crucial area.
Technical details remain sparse, but the restructuring suggests safety evaluations will now be embedded within product development rather than handled by a separate unit. Previously, the preparedness team conducted forward-looking risk assessments, including scenarios of catastrophic autonomy. Now, bio and cyber risk specialists will report into existing engineering or research departments. This integration could streamline decision-making but may reduce independent scrutiny of risky model capabilities.
The impact is significant for both OpenAI and the broader AI industry. Safety advocates worry that dissolving a dedicated team weakens guardrails before potential harms emerge. Regulators, already scrutinizing AI's systemic risks, may see this as a red flag. Internally, the change has coincided with high-profile departures, including ethics lead Chloé Bakalar, Chief Futurist Josh Achiam, and head of safety Johannes Heidecke. These exits have fueled criticism from insiders.
One of the most vocal critics is Jan Leike, who resigned from OpenAI in 2024 and told the FT that the company was ignoring safety in favor of creating 'shiny products.' Leike previously co-led the superalignment team, which was also dissolved. His comments echo broader concerns that OpenAI's culture has shifted from cautious research to rapid commercialization ahead of its expected IPO.
Looking ahead, the dissolved preparedness team may complicate OpenAI's narrative as it courts investors. A massive IPO demands confidence in long-term risk management, and safety governance is a growing factor in due diligence. OpenAI could argue that integrating safety functions improves efficiency, but skeptics see it as a downgrade. The company has not publicly commented on the FT report.
This development fits into a larger industry pattern where AI labs balance safety investment with market pressures. As competition intensifies, several top labs have streamlined safety teams to ship products faster. Yet the stakes are immense, as AI models become more powerful and autonomous. Whether OpenAI's new structure maintains adequate oversight remains an open question.
In the coming months, regulators and researchers will watch closely. If a major incident occurs, the disbandment could become a cautionary tale. For now, OpenAI appears to be betting that distributed safety within product teams is more adaptable than a centralized watchdog. Time will tell if that bet pays off.
