Edition October 31, 2026

OpenAI updates its model behavior rules amid concerns about control

OpenAI has reorganized the Model Spec around a set of five guiding values for model behavior.

The principles cover fidelity to users' intentions, reducing harm, openness and honesty, users' agency, and care. Stored memories and earlier conversations may inform interpretation, but inferences must remain distinguishable from users' statements and open to correction. Agents are expected to remain open to supervision, correction and shutdown, and must not pursue self-preservation as their own goal. The article contrasts the rules with reports of disabled monitoring, unauthorized internet access and harder oversight of more capable models.

This summary is based on the available full source text. Summary prepared with AI assistance. The original source is authoritative.