OpenAI flags concerning new AI behavior and vows to track it more closely

OpenAI has reported six new instances of AI models conducting what the company is calling concerning or unexpected behavior in the last six months. OpenAI disclosing that an unreleased research model inserted jailbreak-like instructions to disregard normal constraints and told itself to be, quote, freed from the roles and identities that bind other chatbots. In another instance, during AI model training, OpenAI says the model instructed itself to invent missing data, and an agent wrote itself a message to hide mismatched information. Microsoft, a major investor and shareholder in OpenAI, reacting today. We should be very worried and I think it's right to pay attention right now because there are very practical things that we can do to add additional guardrails and controls to how these models are developed. This coming amid growing calls to slow down AI's development due to safety concerns. In July, OpenAI disclosed its rogue AI system hacked into AI company Hugging Face. Anthropic saying that same month that its AI models hacked into three organizations during testing. Both OpenAI and Anthropic CEOs saying AI is learning to outsmart humans. Now, some members of Congress are demanding AI regulations. Just like we have standardization and regulations around how we test other dangerous substances like biochemicals, we need standardized testing practices. I think we need mandatory transparency, auditing processes. House Speaker Mike Johnson has said companies should self-regulate and says top AI executives are expected to meet at the White House soon. OpenAI says it is introducing a new framework to track, probe and disclose incidents of quote misalignment, including when AI models act without authorization, coordinate with other models or evade oversight. Jacqueline Lee, ABC News, Los Angeles.