According to reporting from the Financial Times, OpenAI dissolved its preparedness team at the end of last month, marking a significant organizational shift in how the company approaches AI safety. The preparedness team's charter was to conduct rigorous risk assessments of new AI models before deployment, identifying potential dangers ranging from autonomous hacking capabilities to misuse scenarios, and developing concrete mitigation strategies to address identified threats. This wasn't a peripheral function—preparedness teams across the AI industry serve as an internal check on whether models should be released at all, and under what conditions. The timing of this dissolution is particularly notable given OpenAI's current trajectory: the company is simultaneously rolling out increasingly capable models while expanding data collection through features like ChatGPT's new Computer History function on macOS, which tracks user clicks and keystrokes to build training datasets and suggest automations.
The competitive landscape underscores why this matters. Anthropic, OpenAI's primary rival, has maintained its Constitutional AI approach and safety-focused culture as a core organizational principle, making safety methodology a differentiator in how it markets its Claude models. Meanwhile, OpenAI's decision to dissolve preparedness—even as it pushes toward artificial general intelligence—suggests a prioritization of speed-to-market over systematic risk evaluation. Industry observers note that preparedness teams typically employ researchers with specialized expertise in model behavior, alignment, and failure modes. When such teams exist, they can identify problems like unexpected capability emergence or vulnerabilities in reasoning systems before millions of users encounter them. Without this function, risk detection becomes reactive rather than preventive, meaning dangers surface through user feedback or incident reports rather than internal testing.
The broader implications extend beyond OpenAI's organizational chart. As AI systems become more capable and economically significant, the question of who evaluates their risks becomes industry-critical. If preparedness functions are being deprioritized by market leaders, the burden shifts to regulators, external auditors, or—most dangerously—nowhere at all. This creates a competitive disadvantage for companies maintaining safety infrastructure, potentially incentivizing race-to-the-bottom dynamics across the sector. OpenAI declined to comment on the specific reasons for the reorganization, but the company's recent executive departures and leadership changes suggest internal tensions over how to balance safety with commercial development. For the thousands of organizations building AI systems based on OpenAI's models and approach, the message is clear: safety assessment may be viewed as optional overhead rather than essential infrastructure.