OpenAI, the company behind the widely used ChatGPT, has quietly stood up a new internal division whose sole mandate is to prevent the most extreme potential harms of artificial intelligence, including scenarios involving nuclear or biological weapons. The announcement, made via a company blog post, signals a formal acknowledgment of the existential questions that have dogged the industry as AI capabilities accelerate.
The newly formed "preparedness team" is tasked with tracking, evaluating, forecasting, and guarding against a spectrum of threats that could emerge from AI systems. According to the post, these risks span "chemical, biological, radiological, and nuclear" domains, alongside more immediate concerns like AI's ability to manipulate individual beliefs and its role in cybersecurity vulnerabilities.
Notably, the company did not provide a detailed roadmap for how the team will achieve these goals. The blog post offers broad language about protecting against "catastrophic risks" but stops short of outlining specific protocols, tools, or partnerships. This lack of granularity has drawn quiet skepticism from observers who note the tension between OpenAI's profit-driven push to develop ever more advanced models and its stated commitment to safety.
The move comes amid heightened public and regulatory scrutiny of AI's potential downsides. In May, OpenAI CEO Sam Altman testified before Congress, where he candidly acknowledged the stakes. "I think if this technology goes wrong, it can go quite wrong," Altman said at the time. "And we want to be vocal about that. We want to work with the government to prevent that from happening."
Altman's public remarks have often veered between alarm and optimism, a duality that has fueled debate about whether the company's actions align with its rhetoric. The creation of the preparedness team appears to be a direct response to those concerns, though critics argue that building more powerful AI systems is itself a risky gamble.
What the Team Will Actually Do
The preparedness team's mandate is broad, covering both near-term and far-future threats. It will focus on "individual persuasion," a term that suggests addressing AI's capacity to influence people in ways they might not consciously choose. This could encompass everything from targeted disinformation to more subtle forms of behavioral manipulation, though the company has not clarified the boundaries of this work.
Cybersecurity is another pillar of the initiative, but again, specifics are scarce. OpenAI has previously highlighted risks like AI-generated phishing attacks or automated hacking, yet the new announcement offers no concrete strategies for countering such threats.
The team's creation is framed within OpenAI's broader mission of building "safe" artificial general intelligence (AGI)—a hypothetical AI that matches or exceeds human cognitive abilities. Altman has been inconsistent in his public stance on AGI's timeline and dangers, sometimes warning of its risks and other times downplaying them. This inconsistency has led to questions about whether the company's safety commitments are genuine or performative.
Despite the ambiguity, the formation of a dedicated preparedness unit is a notable step for an industry that has often been accused of moving too fast to fully consider consequences. Whether it will meaningfully alter the trajectory of AI development remains an open question, but it does signal that the conversation around AI safety is moving from abstract theory to institutional practice.
