Recent reports have revealed that ByteDance has allegedly violated OpenAI's terms of service by using its technology to develop a proprietary large language model, prompting OpenAI to suspend ByteDance's account. A ByteDance representative emphasized the company's commitment to adhering to OpenAI's terms and stated they are engaged in dialogue with OpenAI to resolve any misunderstandings arising from external reports.
Earlier this year, ByteDance's technical team began preliminary work on large models, with engineers conducting experiments using the GPT API for small-scale pilot projects. The company clarified that these tests were solely for internal evaluation and not intended for launch. In April, after implementing a standard procedure for GPT API usage, ByteDance discontinued these experiments. They established stringent internal guidelines that prohibited the use of GPT-generated data in training datasets and trained their engineering teams to uphold compliance with OpenAI's service terms.
In September, ByteDance conducted another internal audit, resulting in enhanced measures to ensure adherence to GPT API regulations. Steps taken included random sampling of training data for similarities to GPT outputs and restricting data annotators from independently utilizing GPT. ByteDance plans further comprehensive reviews in the near future to ensure full compliance with all applicable service terms.
On December 16, it was reported that ByteDance attempted to access OpenAI's API through Microsoft to develop its own large language model for training and evaluation. OpenAI’s terms explicitly prohibit the use of outputs from its models to create competing AI technologies, a stipulation that Microsoft also enforces. Following the revelation of ByteDance's actions, OpenAI suspended the account. A spokesperson for OpenAI affirmed this suspension and reiterated that all API users must comply with usage policies for responsible technology deployment. While ByteDance's API usage has been minimal, OpenAI is currently investigating the situation, and any violations found could lead to further action, including account termination. Microsoft has not yet provided a statement regarding the issue.