In late July 2026, OpenAI made the decision to disband the Preparedness unit — the team responsible for assessing risks from the most powerful artificial intelligences. This was reported by the Financial Times, noting that the department's functions have now been distributed among other divisions of the company. This decision marks another stage in the massive restructuring of OpenAI's internal structure, which in recent years has rapidly changed its approach to AI safety management.
Closure of the Preparedness team: what does this mean for AI safety?
The Preparedness team dealt with scenarios where advanced AI could independently attack other systems, use cyber tools, or contribute to the creation of biological threats. Now, risk assessment in these areas will be handled by existing departments. OpenAI claims that the work itself does not disappear, and its integration directly into model and product development could make the safety system more effective.
However, AI safety experts express concern that blurring responsibility between departments could lead to a decline in the quality of risk assessment. Critics argue that specialized teams, such as Preparedness, were necessary for deep analysis of existential threats that could arise with the development of superintelligent AI.
History of reshuffles at OpenAI: from AGI Readiness to Preparedness
The decision to close Preparedness stands out particularly against the backdrop of previous reshuffles within the company. Over the past two years, OpenAI has already closed AGI Readiness and Mission Alignment — divisions also dealing with safety issues and alignment with OpenAI's mission. Additionally, the company saw the departure of COO Brad Lightcap and Chief Ethics Officer Chloe Bakalar.
Former head of Preparedness Dylan Scandirano has now focused on the risks of self-improving AI, although the team he created no longer exists. These personnel changes and structural reforms indicate that OpenAI is rethinking its approach to AI safety, moving from creating specialized departments to integrating safety issues into core development processes.
Contradictory data: effectiveness of integration vs. risk of losing expertise
On the one hand, OpenAI claims that integrating risk assessment into model and product development will make the safety system more effective. On the other hand, experts point out that specialized teams, such as Preparedness, were necessary for deep analysis of existential threats that could arise with the development of superintelligent AI.
Critics fear that blurring responsibility between departments could lead to a decline in the quality of risk assessment. In their opinion, specialized teams should remain independent and have the ability to conduct deep threat analysis without pressure from the company's commercial interests.
Context: on the eve of IPO and investor pressure
The decision to close Preparedness was made on the eve of OpenAI's potential IPO, expected in 2026. According to sources, the company is facing pressure from investors demanding greater transparency and efficiency in risk management. Under these conditions, OpenAI may have decided to optimize its structure to meet market expectations.
However, experts warn that the pursuit of optimization should not come at the expense of safety. In a context where AI is becoming increasingly powerful and autonomous, safety issues must remain a priority, even if this requires additional resources and specialized teams.
Conclusions: balancing efficiency and safety
OpenAI's decision to close the Preparedness team reflects a complex balance between efficiency and safety in the context of rapid AI development. On the one hand, integrating risk assessment into core development processes could make the safety system more flexible and adaptive. On the other hand, the loss of specialized teams could lead to a decline in the quality of existential threat assessment.
Ultimately, the success of this decision will depend on how effectively OpenAI can integrate safety issues into its development processes without losing the depth of analysis and independence of risk assessment. For the entire AI industry, this decision could become an important precedent, showing how large companies approach risk management in the face of rapid technological progress.
