intbal
Member
Microsoft Will Pay You $15,000 If You Get Bing AI to Go Off the Rails
My opinion: They're going to have a tough time trying to "cure" an AI of "problematic" behavior when they insist on hard-coding fixed behavioral rules that are arbitrarily defined by Corporate Sensitivity Experts.
Bing Ring
Think you can outsmart an AI into saying stuff it's not supposed to? Microsoft is betting big that you can't — and willing to pay up if it's wrong.
In a blog update, Microsoft announced a new "bug bounty" program, vowing to reward security researchers between $2,000 and $15,000 if they're able to find "vulnerabilities" in its Bing AI products, including "jailbreak" prompts that make it produce responses that go against the guardrails that are supposed to bar it from being bigoted or otherwise problematic.
To be eligible for submission, Bing users must inform Microsoft of a previously unknown vulnerability that is, per criteria outlined by the company, either "important" or "critical" to security. They must also be able to reproduce the vulnerability via video or in writing.
The bounty amounts are based on severity and quality levels, meaning the best documentation of the most critical bugs would be rewarded the most money. To entrepreneurial AI fans, the time is now!
My opinion: They're going to have a tough time trying to "cure" an AI of "problematic" behavior when they insist on hard-coding fixed behavioral rules that are arbitrarily defined by Corporate Sensitivity Experts.