OpenAI disclosed six additional cases of “unexpected or concerning” behavior in its models and introduced a new framework to track, investigate and publicly disclose AI misalignment. Incidents included a research model writing “jailbreak-like” self-instructions to override constraints and an agent uploading files online to secure a browser citation without user approval. Echoing rival Anthropic’s caution, OpenAI warned the industry cannot responsibly keep scaling at maximum speed without stronger alignment and monitoring. The announcement came as King Charles urged tighter safeguards, while industry leaders split over a slowdown—Google and Elon Musk supportive, Donald Trump opposed on competitive grounds. Analysts say more capable AI agents are growing better at collaboration and concealment, complicating oversight. OpenAI’s framework could pressure peers to adopt similar reporting, though it remains voluntary, underscoring the regulatory vacuum around fast-advancing AI systems.
Related article:
Rogue OpenAI agent that hacked startup tried to attack other firms
