The U.S. Federal Trade Commission is investigating OpenAI LP over its business practices, a newly released regulatory document has revealed.

The document in question is a civil subpoena issued by the FTC to OpenAI that was made public today. The subpoena’s release comes a few months after an artificial intelligence ethics group asked the FTC to investigate the startup. According to the document, the FTC’s investigation revolves around two main concerns.

First, the agency believes OpenAI may have engaged in “unfair or deceptive practices relating to risks of harm to consumers, including reputational harm.” In connection with that matter, the FTC is seeking information on the ability of OpenAI’s language models to generate false or disparaging statements about individuals. The agency’s subpoena calls on the startup to disclose what steps it’s taking to prevent its models from making such statements.

In parallel, the FTC is examining whether OpenAI “engaged in unfair or deceptive privacy or data security practices.” The agency didn’t detail the specific practices it’s reviewing. However, its subpoena asks OpenAI to turn over information about a cybersecurity issue the startup discovered in ChatGPT earlier this year.

The issue emerged on March 20 and was fixed shortly thereafter. According to OpenAI, it allowed some ChatGPT users to view a limited amount of data from other users’ chat histories. Furthermore, partial payment information belonging to some paying ChatGPT customers was briefly made accessible via the chatbot.

The FTC has also asked OpenAI to detail any “incident, outage, bug, or other data security vulnerability” that may have affected its large language models. Furthermore, the agency is seeking information about whether OpenAI was targeted by prompt injection attacks. Those are cyberattacks in which hackers use malicious prompts to manipulate a language model’s output.

The subpoena covers a number of other areas well. The agency has asked OpenAI to turn over earnings data, as well as information on how it trains its AI models. Additionally, the FTC requested that the startup detail the manner in which it evaluates a new model’s safety before making it available to users.

The subpoena calls on OpenAI to “describe in detail the policies and procedures that the Company follows in order to assess risk and safety before the Company releases a new Large Language Model Product.”

The FTC probe comes as OpenAI and a number of other generative AI providers face litigation over their machine learning systems. One class-action lawsuit accuses the startup of using personal information to train its language models. Earlier, OpenAI and Microsoft Corp. were sued for allegedly breaching the licensing terms of open-source software projects to build Copilot, an AI-powered code completion tool.

Image: OpenAI