Google's latest AI threat tracker report mostly describes how threat actors are using AI, but the report begins by calling out distillation attacks. Google frames these types of attacks as a form of intellectual property theft.
Google cited one example that involved more than 100,000 queries to its Gemini AI. The campaign appeared to be about "replicating Gemini's reasoning ability in non-English target languages."
Google didn't call out any particular country or competitor, but instead referred to "frequent model extraction attacks from private sector entities all over the world and researchers seeking to clone proprietary logic."
By contrast, in a memo sent to U.S. lawmakers, OpenAI said that the majority of distillation activity appears to originate from China and that it has seen "evolving but persistent methods" being used against its models.
Google's latest AI threat tracker report mostly describes how threat actors are using AI, but the report begins by calling out distillation attacks. Google frames these types of attacks as a form of intellectual property theft.
Google cited one example that involved more than 100,000 queries to its Gemini AI. The campaign appeared to be about "replicating Gemini's reasoning ability in non-English target languages."
Google didn't call out any particular country or competitor, but instead referred to "frequent model extraction attacks from private sector entities all over the world and researchers seeking to clone proprietary logic."
https://news.ycombinator.com/item?id=47106901
By contrast, in a memo sent to U.S. lawmakers, OpenAI said that the majority of distillation activity appears to originate from China and that it has seen "evolving but persistent methods" being used against its models.
https://news.ycombinator.com/item?id=47106746