Microsoft Investigates DeepSeek AI for Potential OpenAI API Misuse

Investigation into DeepSeek's OpenAI API Usage

Following allegations made by David Sacks regarding DeepSeek’s model training practices, Bloomberg Law has reported that Microsoft is currently conducting an investigation. The focus is on DeepSeek’s utilization of OpenAI’s application programming interface (API).

Data Exfiltration Concerns

Microsoft’s security researchers suspect that the Chinese firm responsible for the R1 reasoning model potentially extracted a substantial volume of data through OpenAI’s API during the autumn of 2024. Microsoft, as a major stakeholder in OpenAI, promptly alerted OpenAI to this observed, potentially unauthorized activity.

OpenAI's Terms of Service

Access to OpenAI’s API is generally available to the public. However, OpenAI’s terms of service explicitly prohibit the use of generated output for the purpose of training competing AI models.

The terms clearly state that users are not permitted to “develop models that compete with OpenAI.” Furthermore, automated or programmatic data extraction from the API is also forbidden.

The Role of Distillation

The central concern revolves around the technique of distillation. This method is employed by AI developers to transfer knowledge from one model – the 'teacher' – to another – the 'student'.

It will be crucial to determine if DeepSeek has devised novel strategies to bypass OpenAI’s API rate limitations and execute large-scale queries. Should this be confirmed, legal consequences are highly probable.

Potential Ramifications

• The investigation centers on potential violations of OpenAI’s terms of service.
• Data exfiltration is a key area of concern for Microsoft.
• The use of distillation techniques is under scrutiny.