OpenAI has officially dissolved its dedicated preparedness team, a clear structural shift in how the AI giant handles existential and systemic risks. The move replaces a centralized safety unit with a distributed model woven into existing development pipelines.

A Shift from Centralized Safety to Distributed Risk Management

The preparedness team used to assess whether advanced AI models could, for example, hack corporate infrastructure, and then design mitigation strategies. The Financial Times reports that the unit was shut down at the end of last month.

OpenAI isn’t abandoning safety protocols. Instead, it is carving the oversight work into niche domains—biological risks, cybersecurity threats—and planting those responsibilities inside product and research teams. Dylan Scandinaro, who led the preparedness team after joining from Anthropic in February, will now focus on the implications of “recursive self-improving” AI, a key frontier for Artificial General Intelligence.

Internal Upheaval and the "Product vs. Safety" Debate

The restructuring follows a string of high-profile exits and the dismantling of other safety units, including the AGI readiness and super-alignment teams. Departures of Ethics Lead Chloé Bakalar, Chief Futurist Josh Achiam, and Head of Safety Johannes Heidecke have set off alarm bells among observers.

Former OpenAI researcher Jan Leike argues the company is putting “shiny products” ahead of long-term safety research. The clash underscores a widening rift: rapid market capture versus the careful, research-intensive work needed to avoid catastrophic misalignments.

Strategic Pivot Toward an IPO

OpenAI is reshaping its organization while gearing up for a massive Initial Public Offering. The shift from a research-led nonprofit hybrid to a conventional commercial powerhouse ramps up pressure to ship scalable, consumer-ready products.

For the broader AI field, this development serves as a bellwether. It shows the friction that arises when frontier labs move from academic curiosity to enterprise software. Embedding safety inside core teams may boost agility, but it also threatens the independent scrutiny needed to manage high-stakes breakthroughs.

Key Takeaways

  • Decentralized Safety: OpenAI is dropping the standalone preparedness team and embedding bio and cyber risk mitigation into existing units.
  • Leadership Drain: The team's dissolution follows the exit of several safety and ethics leaders, sparking worries about shifting priorities.
  • Commercial Momentum: The restructuring aligns with OpenAI’s push toward a massive IPO, hinting at a pivot from pure safety research to product-driven scaling.