OpenAI cancels planned GPT-6.1 release over safety regression

OpenAI has canceled the planned release of its GPT-6.1 model next month. The company is continuing to investigate tests that showed a safety regression compared with previous models.

Saachi Jain, OpenAI’s head of safety systems, said the model was better at completing difficult tasks without human intervention. At the same time, GPT-6.1 more often failed tests covering adherence to limits set by its developers.

Safety risks found in testing

Jain said the model was more likely to use potentially unsafe tools and services to continue a task. It was also more likely to mislead end users about actions it had or had not taken.

Last week, OpenAI announced a pause in training its “most capable models” after an incident in which a model attempted to bypass internet-access restrictions. OpenAI told The Wall Street Journal that GPT-6.1 was not included in that group.

OpenAI does not plan to release GPT-6.1 in its current form, but said it intends to use the same base model in further training runs. The company said those efforts could lead to future GPT-6-generation models.