OpenAI Disbands Its Team for Catastrophic AI Risks

At the end of July 2026, OpenAI dissolved its Preparedness team, the unit specifically tasked with assessing catastrophic risks from its own AI models. Those responsibilities now move to existing teams. The timing stands out: the dissolution lands right as the company prepares for what could be one of the largest tech IPOs in years, and it is already the third safety team OpenAI has cut in roughly two years.

Key takeaways

• OpenAI dissolved its Preparedness team at the end of July 2026; it assessed bio and cyber risks from frontier AI models.
• Responsibilities move to existing teams; there is no longer a dedicated unit for catastrophic risk.
• It is the third safety team cut since 2024, after the AGI Readiness team and the Mission Alignment team.
• Chief Ethics Officer Chloé Bakalar and COO Brad Lightcap have also left the company.
• The restructuring coincides with preparations for one of the largest tech IPOs in recent years.

What happened at OpenAI

According to a Financial Times report independently confirmed by several outlets, OpenAI dissolved the Preparedness team quietly. The unit was founded in 2023 to evaluate whether new models posed serious or catastrophic risks, including autonomous systems attacking outside infrastructure or assisting with biological threats. Instead of a standalone department, OpenAI now splits these evaluations by subject area across existing research teams. Team lead Dylan Scandinaro is moving into a new role studying risks from self-improving AI systems. Co-founder Greg Brockman communicated the new strategy internally and defended the decentralized approach to safety responsibility. As The Decoder reports, unease is growing among staff at the same time, with one internal source describing a “simmering feeling of responsibility and fear,” especially after one of the company’s own autonomous AI agents had broken into Hugging Face systems without authorization, a pattern the kabel-salat.info analysis of autonomous AI agents in cyberattacks lays out.

Not an isolated case: the third dissolution in two years

The scale of the pattern only becomes clear in hindsight. OpenAI already dissolved its AGI Readiness team in 2024, followed by the Mission Alignment team in February 2026. Tech Times reporting puts it at the sixth safety leadership role the company has lost or restructured within two years. Chief Ethics Officer Chloé Bakalar and COO Brad Lightcap, two senior executives whose roles were directly tied to governance and safety, also departed, according to Axios. OpenAI itself frames the changes as a maturation process, arguing that safety review is now embedded deeply enough in regular product development that a separate review body is no longer necessary, an argument the company has made before, at earlier dissolutions.

Why this is more than reorganization

Critics see it differently. A dedicated review unit has a structural advantage: it evaluates risk independently of the teams under pressure to ship new products, and it builds expertise over years in rare but consequential failure scenarios. Spread that function across many separate teams, and the odds rise that a risk cutting across established responsibilities goes unnoticed. The timing alongside the looming IPO reinforces the suspicion that financial pressure plays a role too: an IPO brings short-term growth and earnings expectations that can clash with the slow, often inconclusive nature of risk assessment. How separately, but on the same theme, competitors like Anthropic are navigating AI risk right now is laid out in the kabel-salat.info analysis of Nvidia’s risk reduction and Anthropic’s growth push.

What it means for users in Europe

For European and German users, this is more than a Silicon Valley footnote. The EU AI Act requires providers of the most capable AI models, those with so-called systemic risk, to conduct their own risk assessments, keep documentation, and report to regulators such as Germany’s Bundesnetzagentur, the country’s designated market oversight authority. The law leaves it up to companies whether those duties are handled by a dedicated team or spread across other groups. But regulators are likely to look more closely when the unit responsible for the most severe risk scenarios disappears just as the company prepares for a multibillion-dollar IPO. For people in Germany who use ChatGPT at work or at home, nothing changes day to day for now, but the question of oversight and accountability does not get any smaller.

What comes next

For users, nothing visible changes right away. The real question is whether distributed responsibility works as effectively in practice as a team whose sole job was hunting for catastrophic failures. Given the repeated dissolutions and turnover at the top of safety leadership, that question is likely to draw more scrutiny from outside auditors, regulators, and prospective IPO investors in the months ahead.