ChatGPT, arguably the most well-liked OpenAI product, is on the middle of the bug bounty program. The corporate says it presents rewards for locating flaws related to the ChatGPT Plus subscription service, person log-in, and different associated functionalities. ChatGPT was not too long ago pulled down briefly following a bug that uncovered private data tied to the subscription, reminiscent of actual identify and banking particulars. OpenAI will even provide a bounty to consultants that uncover flaws in the freshly-launched plug-in system, which permits shoppers like Expedia to combine the pure language bot into their on-line providers.
Notably, the bug bounty does not cowl flaws related to the text-based prompts given to ChatGPT. OpenAI notes on the BugCrowd listing that “points associated to the content material of mannequin prompts and responses are strictly out of scope, and won’t be rewarded.”
Weeks in the past, some testers had been in a position to bypass ChatGPT’s delicate content material restrictions utilizing the immediate system with out even having to tinker with any code-level settings. The difficulty acquired widespread consideration, however surprisingly, OpenAI will not take into account prompt-related flaws bounty-worthy until they pose a direct safety menace.