Reported by cyberscoop. A new bipartisan House bill aims to protect American artificial intelligence models by tasking the federal government with creating formal security guidance against distillation attacks.
The AI Cyber Defense Act, sponsored by Rep. Pat Harrigan, a Republican from North Carolina, and Rep. George Whitesides, a Democrat from California, directs the National Institute for Standards and Technology to establish voluntary industry standards. These measures seek to prevent foreign entities in Russia, China, Iran and North Korea from using distillation attacks to copy U.S. technology. Distillation involves training an AI model on the outputs of a more advanced system, allowing adversaries to replicate capabilities at a fraction of the original development cost. U.S. officials and companies like Anthropic, OpenAI and Google have identified systematic campaigns where actors use burner accounts to extract data from frontier models. The legislation also creates an AI Cybersecurity Roundtable at the National Institute for Standards and Technology to foster collaboration between government and private stakeholders. This initiative aims to secure critical infrastructure while maintaining the 6-to-12 month competitive lead held by the U.S. AI industry. By defining clear defensive standards, the bill intends to mitigate the risk of foreign actors leapfrogging American innovation through intellectual property theft.
Source: cyberscoop
