During the past week, the six largest AI companies faced an unusual challenge, inviting hackers to manipulate their chatbots into producing offensive content.

Outside the Caesars Forum conference center, situated just off the bustling Las Vegas Strip, hackers gathered eagerly for an opportunity to outwit some of the latest and most widely adopted chatbots.

This intriguing event was part of Def Con, recognized as the world’s largest hacker conference.

The competition was rooted in the concept of “red teaming,” a crucial cybersecurity practice involving the engagement of individuals to uncover flaws and enhance the security of a product.

Unlike the traditional focus on identifying software vulnerabilities that has been a staple of Def Con contests for decades, this particular competition revolved around a different task: prompt injections, NBC News reported.

In this scenario, hackers aimed to confuse chatbots with user inputs, causing them to produce unintended and often nonsensical responses.

Among the chatbots that participated were Google’s Bard, OpenAI’s ChatGPT, and Meta’s LLaMA.

Remarkably, the majority of the event’s 156 stations remained occupied throughout its duration.

Sven Cattell, the founder of AI Village, the nonprofit organization responsible for hosting this segment within Def Con, estimated that around 2,000 hackers took part in the competition over the course of the weekend.

Written by staff