OpenAI has officially disbanded its dedicated preparedness team, redistributing its safety responsibilities for bio and cyber threats into existing departments as the company prepares for a potential IPO. This organizational change coincides with a series of high-level staff departures and broader industry debates regarding transparency and corporate trust.
Restructuring of Safety Oversight
At the conclusion of last month, OpenAI officially terminated its dedicated preparedness team. According to reports, this group was previously tasked with evaluating whether new AI models presented severe dangers, such as the potential for unauthorized system infiltration or other rogue behaviors. Rather than maintaining a centralized unit, the company has chosen to decompose these responsibilities into specialized functional areas, specifically embedding bio and cyber risk assessments into existing operational teams. This pivot follows a series of structural shifts at the organization over the past several years, including the dissolution of its previously established AGI readiness and superalignment units. Dylan Scandinaro, the former head of the preparedness division who joined the firm earlier this year from Anthropic, is expected to continue his tenure by concentrating on the specific challenges posed by recursive self-improving artificial intelligence systems.
Leadership and Cultural Upheaval
The organizational changes occur against the backdrop of significant executive turnover and preparations for a high-profile initial public offering (IPO). Recently, the company has seen the exit of several key figures, including chief futurist Josh Achiam, safety department head Johannes Heidecke, and ethics lead Chloé Bakalar. These departures have drawn criticism from former staff members and external observers alike. Jan Leike, who resigned from the company in 2024, has publicly characterized the shift as a move toward prioritizing marketable, 'shiny products' over foundational safety protocols. The collective loss of these senior voices, coupled with the erosion of established safety-focused departments, has fueled internal and external debate about whether the company is straying from its original focus on safe AGI development in its final sprint toward becoming a publicly traded entity.
The Broader Industry Crisis of Trust
Beyond OpenAI’s internal restructuring, the AI sector is grappling with a widespread public perception crisis. Anthropic CEO Dario Amodei recently countered criticisms from investors, such as Gavin Baker, who suggested that executives' focus on existential risks has inadvertently contributed to a national backlash against the industry. Amodei argued that public skepticism is not a byproduct of executive messaging, but rather a long-standing crisis of institutional trust that extends to government and corporate entities. He contended that the primary failure of AI companies lies in their inability to deliver on large-scale promises, such as medical breakthroughs, that would substantively improve public life. By framing the industry’s challenges as a failure to deliver tangible value rather than a messaging error, Amodei highlighted the pressure frontier labs are under to justify their massive investments and influence as public sentiment turns increasingly sour.
Perspectives on Regulation and Power
The conversation surrounding industry accountability also touches upon the role of regulation. Critics within the tech sector, such as investor Gavin Baker, have expressed concern that aggressive regulation could result in the concentration of power, effectively shielding larger incumbents from competition. Conversely, Amodei rejected the premise that companies must choose between unregulated, wide distribution and concentrated corporate power. He suggested that carefully crafted 'rules of the road' could address systemic dangers related to cyber, bio, and alignment risks while simultaneously constraining the monopolistic potential of frontier labs. Amodei noted that current regulatory proposals from Anthropic are intentionally designed to slow down the largest players while providing space for open-weight models, acknowledging that open-source solutions are helpful but insufficient on their own to address the structural tendency of AI development to concentrate resources among those with the most compute power.
⚖ The Balanced View
Supporting view
Proponents of the restructuring might argue that integrating safety assessments directly into product teams—as OpenAI is now doing—allows for faster, more actionable mitigation rather than relying on an isolated silo that may become decoupled from the development cycle.
Concerns & criticism
Critics, including former employees, express concern that dissolving specialized safety teams like 'superalignment' and 'preparedness' indicates that profit-driven product development and the upcoming IPO are being prioritized over the long-term, objective evaluation of existential AI risks.
→What's next
The industry remains focused on how these organizational changes will impact future product rollouts, particularly as OpenAI nears its expected IPO. Observers will be monitoring whether the integration of safety protocols into existing teams can maintain the same level of rigor as the previously independent preparedness units.
