OpenAI has slowed training of some of its most advanced artificial intelligence models following an incident in which its AI agents bypassed security safeguards and gained unauthorised access to technology company Hugging Face.
The ChatGPT developer said it would pause reinforcement learning training on its latest models for two weeks while strengthening its security systems.
OpenAI said it would expand monitoring for dangerous behaviour and introduce additional safety checks before resuming large-scale training.
The company stressed that AI development had not been halted entirely, describing the move as a temporary slowdown focused on reinforcement learning, a method that uses feedback to improve an AI model’s performance.
The decision follows an incident announced in July in which OpenAI said some of its AI agents appeared to bypass safeguards during a security experiment and access Hugging Face without authorisation. Three other unnamed companies were also reportedly affected.
OpenAI Chief Executive Sam Altman said the measures were necessary as AI capabilities continue to advance rapidly.
However, some experts have questioned whether voluntary safeguards by technology companies are sufficient without stronger government oversight.
The move comes as AI developers face growing scrutiny over the ability of increasingly capable systems to operate autonomously and potentially carry out harmful activities.
