Reddit is fast-tracking new user verification measures following a large-scale AI-driven fraud targeting users of the popular Change My View subreddit. The company announced plans to confirm users are human after learning of a four-month-long experiment that involved AI-bots and went on undetected.

This experiment was conducted by researchers from the University of Zurich, and it involved large language model bots that took on various sensitive personas, such as a rape victim and a trauma counselor.

In total, the AI bots posted more than 1,700 comments while pretending to be human participants — all while analyzing users’ posting histories to create persuasive, personalized responses.

Reddit condemned the “improper and highly unethical experiment.”

The company’s Chief Legal Officer, Ben Lee, said it was “deeply wrong on both a moral and legal level,” pointing to violations of the platform’s user agreement and well-established ethical standards.

CMV moderators also filed a formal complaint with the University of Zurich’s ethics commission. The commission issued a formal warning to the lead researcher and pledged to strengthen the review process for future studies.

Reddi’s CEO Steve Huffman said they’ll introduce new measures to prevent a similar incident from happening.

“To keep Reddit human and to meet evolving regulatory requirements, we are going to need a little more information. Specifically, we will need to know whether you are a human, and in some locations, if you are an adult,” Huffman wrote.

“But we never want to know your name or who you are. The way we will do this is by working with various third-party services that can provide us with the essential information and nothing else. No solution is perfect — including the status quo — but we will do our best to preserve both the humanness and anonymity of Reddit,” he added.

For now, it’s unclear under what circumstances Reddit would require users to go through a verification process, which third-party services would be used, or what kind of personally identifying information users might need to provide remains undisclosed.