OpenAI Disbands Its Risk‑Assessment Team: What It Means for AI Safety

According to a report by the Financial Times, OpenAI shut down its “preparedness” team at the end of last month. The unit was tasked with evaluating whether the company’s AI models posed serious threats and designing mitigation strategies—ranging from accidental bias to the more dramatic scenario of a model turning rogue and hacking a corporate network.

OpenAI did not issue a public statement about the closure; the news surfaced through internal sources that were later confirmed by the FT. Company officials say that risk assessment will no longer live in a siloed department but will be woven into broader security, policy, and ethics groups across the organization. The goal, according to insiders, is to make safety checks faster, more adaptable, and better aligned with product development cycles.

Analysts see two possible interpretations. On one hand, eliminating a dedicated risk‑focused team could signal a deprioritisation of safety, especially as regulators worldwide tighten scrutiny over powerful AI systems. On the other hand, integrating risk work into multiple functions might lead to more holistic oversight, preventing the bottlenecks that a single team can create.

Financial analysts also note that the move comes amid rising investor pressure for transparent governance and compliance with emerging AI regulations. OpenAI is reportedly drafting a new “responsibility framework” that will cover model deployment, continuous monitoring, and emergency response protocols.

Industry experts say they will watch closely how OpenAI operationalises this framework. Many are calling for a public, auditable reporting mechanism that details how large language models are tested against rogue‑behaviour scenarios. The decision to dissolve the preparedness team could become a case study in how leading AI firms balance rapid innovation with the growing demand for robust safety standards.

Source: The Verge

ETİKETLER