OpenAI Dissolves Its Risk Preparedness Team Ahead of IPO

OpenAI is undergoing a series of substantial changes to its internal structure, amid preparations for an initial public offering expected to be among the largest in the technology sector. The latest of these changes, according to a report by the Financial Times, was the dissolution of the Preparedness Team at the end of last month, a team that carried out a sensitive mission related to assessing the potentially serious risks of intelligent models.

What Role Did the Preparedness Team Play?

This team was responsible for studying the dangerous possibilities that could arise from the development of AI models, and for devising mechanisms to mitigate them before they turned into actual threats. Among the scenarios it worked on evaluating was the possibility of models going out of control or being used in malicious activities such as breaching other companies' systems.

According to the report, these responsibilities were not entirely eliminated, but rather distributed across specialized areas such as biological and cyber risks, and transferred to existing teams within the company instead of keeping them within a single independent entity.

A Series of Dismantlements in Safety Teams

This change is not an isolated event, but rather comes as an extension of a gradual trajectory the company has witnessed in recent years, during which it moved away from its model that was more focused on the long-term dimensions of safety. OpenAI had previously dissolved important teams, including:

• The AGI Readiness team.
• The Superalignment team, concerned with ensuring that advanced models align with human values.

This was accompanied by a wave of departures among leaders concerned with ethics and safety, including ethics lead Chloe Bakalar, chief futures expert Josh Achiam, and safety lead Johannes Heidecke.

Rising Criticism Over the Company's Priorities

These changes sparked a wave of criticism from within the tech community and beyond. Among the most prominent critics is Jan Leike, who resigned from OpenAI in 2024, as he believed the company had come to neglect safety aspects in favor of focusing on launching "shiny products" that serve its commercial goals.

These concerns reflect a broader debate in the AI industry about the balance between the speed of innovation and commercial release on one hand, and commitment to safety controls and responsibility on the other, especially with the accelerating development of the capabilities possessed by modern models.

A New Direction Toward Recursive Self-Improvement

In a related context, the report indicated that Dylan Scandinaro, who led the Preparedness Team after joining the company from Anthropic last February, will now focus on studying the implications of what is known as "recursive self-improvement" of AI; a concept that refers to the ability of models to develop themselves at an accelerating pace without direct human intervention, and one of the most hotly debated topics among researchers.

Conclusion

These developments show that OpenAI is going through a deep restructuring phase ahead of its IPO, in which it is reordering its priorities and redistributing its safety-related responsibilities. While the company asserts that these steps do not mean abandoning safety but rather reintegrating it within work teams, the question remains about the extent of the company's ability to balance its commercial ambitions with its commitments toward the long-term risks of the technologies it develops.

✦ بقلم فريق دروب أيديا