[ad_1]
Google has introduced that it is increasing its Vulnerability Rewards Program (VRP) to compensate researchers for locating assault situations tailor-made to generative synthetic intelligence (AI) methods in an effort to bolster AI security and safety.
“Generative AI raises new and totally different considerations than conventional digital safety, such because the potential for unfair bias, mannequin manipulation or misinterpretations of knowledge (hallucinations),” Google’s Laurie Richardson and Royal Hansen mentioned.
A few of the classes which might be in scope embody immediate injections, leakage of delicate information from coaching datasets, mannequin manipulation, adversarial perturbation assaults that set off misclassification, and mannequin theft.
It is price noting that Google earlier this July instituted an AI Crimson Workforce to assist deal with threats to AI methods as a part of its Safe AI Framework (SAIF).
Additionally introduced as a part of its dedication to safe AI are efforts to strengthen the AI provide chain by way of current open-source safety initiatives corresponding to Provide Chain Ranges for Software program Artifacts (SLSA) and Sigstore.
“Digital signatures, corresponding to these from Sigstore, which permit customers to confirm that the software program wasn’t tampered with or changed,” Google mentioned.
“Metadata corresponding to SLSA provenance that inform us what’s in software program and the way it was constructed, permitting shoppers to make sure license compatibility, establish recognized vulnerabilities, and detect extra superior threats.”
The event comes as OpenAI unveiled a brand new inside Preparedness staff to “monitor, consider, forecast, and defend” in opposition to catastrophic dangers to generative AI spanning cybersecurity, chemical, organic, radiological, and nuclear (CBRN) threats.
The 2 corporations, alongside Anthropic and Microsoft, have additionally introduced the creation of a $10 million AI Security Fund, targeted on selling analysis within the discipline of AI security.
[ad_2]