OpenAI Stung by Israeli Startup's Gambling Bot Army Spamming X Users

OpenAI has pulled the plug on a sprawling spam network it’s calling “Bet Bot,” an operation that used its models through an Israel-based startup to run a grubby gambling link hustle on X. The scheme wasn’t about politics or ideology. It was pure, old-fashioned spam dressed up in a new AI wardrobe. Operators had the models craft replies and direct messages designed to slip links to online gambling sites into unsuspecting users’ inboxes.

The level of tradecraft was a mixed bag. On one hand, the network was sophisticated enough to manage a fleet of fake personas with AI-generated bios and profile pictures, often themed around UK soccer with bios claiming roots in Manchester or Liverpool. The bots would scan X for conversations that already had engagement and inject a seemingly relevant comment, using OpenAI’s API to draft the text. But the illusion often had cracks. Some accounts reused supposedly unique AI profile pictures, and one persona went by the glaringly fake name “KobeBryantJohnson.” It’s a reminder that even in automated disinformation, human laziness is a constant.

Publicly, the fake accounts masked their true purpose by chatting about sports—football, soccer, baseball—and occasionally throwing in some apolitical, scattershot commentary on US and UK politics to look like real people. The real payload was reserved for direct messages. In those private channels, the conversation pivoted sharply to touting gambling sites via shortened bit.ly links. While OpenAI can’t confirm how many DMs were actually sent, it says some prompts suggested multi-exchange conversations were happening. That broke the bots out of their own sterile echo chamber, a key metric that landed this a Category 2 on its Breakout Scale.

The real takeaway for the industry is less about the specific tactic and more about the target. This wasn’t a state actor or a propaganda mill. It was a straightforward commercial scam exploiting a cost arbitrage in generative AI. The fact that the actors found a way to push links into private DMs through X is a practical problem for platform security teams far more than it is a high-concept AI safety threat. Still, with most fake accounts struggling to attract more than a handful of followers, the operation’s effectiveness seems dubious. As one prompt suggested, people who were messaged often shot back that they don’t gamble online. The bots were built, but the marks weren’t biting.