OpenAI is introducing a special cash payout program for people who report security flaws, vulnerabilities, or other problems in AI systems.
Both security experts and ordinary people who like to explore technology can participate in the program. For a mistake, they can pay from $200 to $20,000 – depending on the “seriousness” of the find. Transactions and payouts will be handled by the Bugcrowd platform.
Similar programs have already been implemented by major technology companies such as Google and Apple. The first in 2019 paid a total of $6.5 million to people for found errors (the reward for one reached $201,337). While Apple went further and offered up to $2 million to anyone who discovers “problems that bypass special lock mode protection.”
ChatGPT ran into privacy issues last month when it accidentally “lit up third-party chat headers.” A few days after that, a Twitter user wrote that he found more than 80 secret plugins during the chatbot hack.
This evening I was disabled the new ChatGPT API and appeared some super interesting: there are more than 80 secret plugins that can be updated to change a specific parameter from the API Call.
Additional modules include “DAN plugin”, “Crypto Prices Plugin”, and many others. pic.twitter.com/Q6JO1VLz5x
— ?? (@rez0__) March 24, 2023
However, jailbreaks and incorrect or negative ChatGPT responses are unlikely to be rewarded. With its announcement, OpenAI tries to say that it cares about privacy and security, but also clarifies that it “cannot predict how people will use these technologies in the real world.”