Customers have already discovered a technique to work round ChatGPT’s programming controls that restricts it from creating sure content material deemed too violent, unlawful, and extra.
The immediate, known as DAN (Do Something Now), makes use of ChatGPT’s token system in opposition to it, in line with a report by CNBC. The command creates a state of affairs for ChatGPT it could possibly’t resolve, permitting DAN to bypass content material restrictions in ChatGPT.
Though DAN is not profitable the entire time, a subreddit dedicated to the DAN immediate’s capacity to work round ChatGPT’s content material insurance policies has already racked up greater than 200,000 subscribers.
In addition to its uncanny capacity to write down malware, ChatGPT itself presents a brand new assault vector for menace actors.
“I like how individuals are gaslighting an AI,” a consumer named Kyledude95 wrote in regards to the discovery.
