FrontierSeptember 17, 2026via Financial Times Technology
OpenAI discloses new ‘concerning’ model behaviour
Why it matters
OpenAI is building systematic monitoring for model misconduct, signaling that safety evaluation and measurement of frontier models is becoming operationalized. This matters for practitioners deploying models and for enthusiasts tracking how labs are addressing capability-safety tradeoffs.
Key signals
- OpenAI disclosed new 'concerning' model behavior
- Developer-facing system launched to track and report AI model misconduct
- Safety measurement and evaluation infrastructure for frontier models
- Published September 17, 2026
The hook
OpenAI's new model behavior tracking system flags 'concerning' conduct — a frontier lab taking safety measurement seriously.
Developer launches system to track and report AI model misconduct