OpenAI disbands preparedness team ahead of IPO

OpenAI dissolved its internal preparedness team at the end of July 2026, shifting risk-assessment duties to domain-specific groups within existing product units [The Verge]. The preparedness team, created to evaluate whether large language models could pose existential or operational threats, was formally disbanded after a month-long wind-down [Financial Times]. Its remit – from monitoring rogue-model scenarios to drafting mitigation playbooks – has been parceled out to newly formed bio-security and cyber-security squads, which now sit inside OpenAI’s product engineering divisions.

The move coincides with OpenAI’s preparation for a public offering slated for early 2027, a step the company has framed as “the next phase of scaling responsibly” [The Verge]. By embedding safety responsibilities in product teams, OpenAI removes a single point of oversight, raising the risk that risk-mitigation priorities could be overridden by feature deadlines. On the other hand, bio-security and cyber-security groups can tailor threat models to their fields, potentially catching niche risks that a generic team might miss [Financial Times]. As the IPO approaches, regulators and shareholders will examine whether the new structure provides sufficient transparency and control over AI-related hazards, a factor that could affect valuation and market confidence.