OpenAI Says It Disrupted a Coordinated Model-Distillation Campaign

It is an attempt to extract the behavior or reasoning of a closed model by sending large volumes of queries to it and then training another model on its outputs, often without the original provider's permission.

Because many regional teams build custom models on top of APIs from global providers, exposing them to legal and technical risk if they go beyond the provider's terms of service.

It depends on the terms of service and licenses. Some providers prohibit using outputs to train competing models, while others allow it within limits. Teams should review the terms before relying on those outputs.