OpenAI Disbands Its Preparedness Team Amid Safety Shakeup

OpenAI has reportedly dissolved its internal preparedness team, according to a report from the Financial Times. The specialized unit, which was tasked with evaluating advanced models for severe risks such as cyberattacks and biological threats, was disbanded at the end of last month. Rather than maintaining a centralized group, OpenAI has distributed these responsibilities across existing teams, dividing the oversight of specific domains like biosecurity and cybersecurity.

This restructuring represents the latest in a series of organizational shifts at the San Francisco-based startup, which is currently restructuring its corporate setup ahead of a highly anticipated initial public offering. Over the past year, OpenAI has dismantled several other safety-focused groups, including its superalignment and artificial general intelligence readiness teams. The company has also seen a wave of high-profile departures, including ethics lead Chloé Bakalar, chief futurist Josh Achiam, and safety head Johannes Heidecke. Former alignment lead Jan Leike, who resigned earlier this year, previously criticized the firm for prioritizing commercialized products over safety.

Dylan Scandinaro, who was hired from rival AI lab Anthropic in February to lead the preparedness team, will remain at OpenAI. However, his focus will shift to studying the implications of 'recursive self-improving' artificial intelligence. For developers and enterprise practitioners, these structural changes signal a shift in how OpenAI manages safety protocols. Instead of a dedicated, independent gatekeeper evaluating catastrophic risks, safety evaluations are now integrated directly into broader product development pipelines. This decentralized approach could accelerate product deployment but may raise questions about the independence of safety audits.