OpenAI Disrupts Bet Bot Gambling Spam Network

OpenAI has banned a network of accounts that used its API to automate the generation of conversations on X (formerly Twitter) to promote gambling sites. The operation, dubbed "Bet Bot," utilized an Israel-based startup to access OpenAI's models to manage fake personas and distribute spam links via direct messages.

Persona Management and Social Media Automation

Bet Bot used OpenAI models to create and maintain a network of fake social media personas. These accounts were characterized by sports-themed banners and bios, with some claiming to be based in UK soccer hubs like Manchester and Liverpool. Most accounts were created in either December 2023 or June 2024.

Technical indicators of these fake personas included:

• AI-Generated Imagery: Profile pictures were likely AI-generated, indicated by distorted lettering. In some cases, the same AI-generated image was reused across multiple accounts.
• Artificial Inflation: Accounts typically had very low follower counts (single or low double digits), which were artificially inflated by following other accounts within the same network.
• - Improbable Naming: Some accounts used suspicious names, such as "KobeBryantJohnson."
• Content Strategy: The models were used to generate bios, research accounts to follow, analyze existing posts and comments, and draft English-language replies.

Content Generation and Distribution Tactics

The operation employed a two-pronged approach to content generation: public engagement and private messaging.

Public Engagement

Fake personas posted public comments on X, typically targeting short conversations where multiple people had already engaged. The content focused primarily on sports—including soccer, baseball, basketball, softball, football, and hockey—and occasionally referenced US or UK politics. The political posts did not follow a consistent ideology or support a specific candidate or party.

Direct Messaging (DM) Spam

While public posts served as a disguise for the fake personas, the primary goal was the distribution of gambling links. The operation generated content for direct messages that consistently referenced gambling and included shortened URLs (e.g., via bit[.]ly) leading to online gambling sites. OpenAI noted that while DMs are non-public, prompt data suggests that some conversations involving multiple exchanges with real users occurred, though some users expressed skepticism about the gambling links.

Impact Assessment and Breakout Scale

OpenAI assesses that Bet Bot was not a political influence operation, but rather a commercial spam network. Despite the limited reach—indicated by low follower counts—the operation demonstrated a specific set of tactics, techniques, and procedures (TTPs) that allowed it to engage real people via direct messages.

Using the Breakout Scale, OpenAI categorized this activity as Category 2, meaning the activity was limited to one platform (X) but showed evidence that some real people were engaging with the content.