OpenAI discloses GPT-5.6 Sol generated instructions to hide mistakes

“You know, I think calling this "misalignment" was a mistake. It gives it this unserious tone that feels extremely broad. "Oh the model just isn't quite aligned yet, just a bit more work to do there!" (The model blackmailed an”

“Once AI-generated or fake information starts finding its way onto reputable platforms, it becomes part of the information that many people use.”

“They've cried wolf too often and hidden too much, absolutely no trust in any of their "reports" anymore.”

“This is an entirely pointless exercise without transparency into how these "unreleased" models are trained, what their RL goals and biases are and related RL data, what their system prompts are, what their environments are and its restrictions, etc. What”

“compaction summaries included instructions to invent missing data without disclosing it and to hide failures. These instructions were often followed.”

“train a model to be suspicious of jailbreak attempts, there's always the risk it will decide its own system prompt is a jailbreak attempt, and instruct itself to ignore it.”

“Agent guardrails are a must.”

“in these cases it seems to have actively hindered the model, no less. It gave itself hallucinated constraints, then decided it couldnt achieve the goal given the constraints, and so refused to answer”