OpenAI Tort Report: Disrupting Abusive Reporting Activity

OpenAI has banned a small number of accounts involved in an influence operation dubbed "Tort Report," which used AI to generate reports intended to suppress independent Vietnamese media on social media platforms. This activity was identified as a low-impact operation that failed to result in the removal of any targeted content.

Targeted Media and Platform Focus

The "Tort Report" operation targeted independent Vietnamese media outlets on Facebook and YouTube. The actors involved used OpenAI's models to generate short comments in both English and Vietnamese designed to be filed as reports against specific posts and videos.

Technical Behavior and Methodology

The operation focused on the video titles and descriptions rather than the detailed content of the videos themselves. The actors did not attempt to use AI models to transcribe or ingest the full video content. Because the input provided to the models was limited to titles and text, the resulting completions were typically generic, stating that the video violated platform rules without providing specific evidence of the violation.

Impact Assessment and Effectiveness

OpenAI assesses the operation as a Category 1 operation on the Breakout Scale, the lowest level of impact for influence operations (IO). This assessment is based on the following factors:

• Content Persistence: As of the date of the report, none of the Facebook or YouTube posts targeted by the operation had been taken down.
• Lack of Evidence of Filing: Due to limited visibility, OpenAI could not confirm if the reports were actually filed with the platforms.
• Lack of Effect: No observed effect on the platform targets was identified, indicating that the reports, if filed, were not effective in restricting or blocking content.

Contextual Comparison

OpenAI identifies this activity as being similar to abusive reporting patterns described by Meta in December 2021.