\n<\/p>\n
OpenAI has launched a bug bounty<\/a>, encouraging members of the public to find and disclose vulnerabilities in its AI services including ChatGPT. Rewards range from $200 for \u201clow-severity findings\u201d to $20,000 for \u201cexceptional discoveries,\u201d and reports are submittable via crowdsourcing cybersecurity platform Bugcrowd<\/a>.<\/p>\n<\/div>\n Notably, the bounty excludes rewards for jailbreaking ChatGPT or causing it to generate malicious code or text. \u201cIssues related to the content of model prompts and responses are strictly out of scope, and will not be rewarded,\u201d says OpenAI\u2019s Bugcrowd page<\/a>.<\/p>\n<\/div>\n Jailbreaking ChatGPT usually involves inputting elaborate scenarios in the system that allow it to bypass its own safety filters. These might include encouraging the chatbot to roleplay as its \u201cevil twin,\u201d letting the user elicit otherwise banned responses, like hate speech or instructions for making weapons. <\/p>\n<\/div>\n OpenAI says that such \u201cmodel safety issues do not fit well within a bug bounty program, as they are not individual, discrete bugs that can be directly fixed.\u201d The company notes that \u201caddressing these issues often involves substantial research and a broader approach\u201d and reports for such problems should be submitted via the company\u2019s model feedback page<\/a>.<\/p>\n<\/div>\n Although such jailbreaks demonstrate the wider vulnerabilities of AI systems, they are likely less of a problem directly for OpenAI compared to traditional security failures. For example, last month, a hacker known as rez0 was able to reveal 80 \u201csecret plugins<\/a>\u201d for the ChatGPT API \u2014\u00a0as-yet-unreleased or experimental add-ons for the company\u2019s chatbot. (Rez0 noted that the vulnerability was patched within a day after they disclosed it on Twitter.)<\/p>\n<\/div>\n