In light of the unique security issues posed by generative AI, such as potential unfair bias or model manipulation, Google is reevaluating how it categorizes and reports bugs. This is being done with insights from its newly established AI Red Team, a group of hackers simulating various adversaries to identify security weaknesses in technology. The team recently conducted an exercise to identify the most significant threats to the technology behind generative AI products like ChatGPT and Google Bard.
Key takeaways:
- Google has expanded its vulnerability rewards program (VRP) to include attack scenarios specific to generative AI.
- The expansion of the VRP is aimed at incentivizing research around AI safety and security, and identifying potential issues to make AI safer for everyone.
- Google is rethinking how bugs it receives should be categorized and reported, given the new security issues brought by generative AI, such as potential for unfair bias or model manipulation.
- Google is using findings from its newly formed AI Red Team, a group of hackers simulating various adversaries, to hunt down security weaknesses in technology, particularly in generative AI products like ChatGPT and Google Bard.