Amazon grew to become the most recent firm to open its giant language fashions to exterior safety researchers, announcing the creation of a brand new bug bounty program for the tech large’s AI instruments.
This system will permit choose third-party researchers and tutorial groups to prod NOVA, Amazon’s suite of foundational AI fashions and obtain compensation for his or her findings. It’ll cowl a spread of common vulnerabilities that have an effect on most generative AI techniques: immediate injection, jailbreaking and vulnerabilities inside the mannequin which have “real-world exploitation potential.” Researchers can even take a look at how the fashions might be manipulated to assist in the production of chemical, organic, radiological and nuclear weapons.
“Safety researchers are the final word real-world validators that our AI fashions and functions are holding up underneath artistic scrutiny,” Hudson Thrift, CISO of Amazon Shops, stated in an announcement Tuesday.
Individuals might be chosen subsequent yr by Amazon…
