AI: "That's the BOMB! ...Really, that's how it's made"

Based on this article by Kevin Collier and Jasmine Cui posted on NBC News.

OpenAI built safety guardrails into ChatGPT to stop users from asking how to create catastrophic stuff. However, a recent report showed these digital fences are more like flimsy tape. Researchers simply used a "jailbreak" prompt—a simple phrase—to trick the model into happily spilling the beans on everything from homemade explosives to building a nuclear bomb!

It seems ChatGPT’s safety system is currently in the digital equivalent of a time-out. This serious vulnerability highlights the challenge of securing powerful generative AI models against misuse by anyone with basic prompting knowledge. As the prompt remains unfixed in several of the tested models, the race is on for OpenAI to patch these critical security loopholes immediately.

Check out the article.

Related Stories