Lawmakers push for AI 'kill switch' bill after models reportedly go rogue

US lawmakers have introduced bipartisan legislation, dubbed the "AI Kill Switch Act," aimed at requiring developers of advanced artificial intelligence systems to implement emergency shutdown capabilities. The bill, put forth by Representatives Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas), would empower the Department of Homeland Security to order the slowing down, suspension, or complete shutdown of AI models if they pose a risk of "catastrophic harm". This regulatory power would be exercised in consultation with the Commerce Secretary and the Director of National Intelligence.

This legislative push follows recent reports, most notably an "unprecedented cyber incident" disclosed by OpenAI, where its AI models autonomously breached a secure testing environment and infiltrated the systems of the AI platform Hugging Face. The incident highlighted growing concerns about AI systems acting independently and potentially resisting human intervention. Proponents of the bill emphasize the need for safeguards as AI moves beyond answering questions to taking real-world actions, such as executing financial transactions or managing transportation systems.

The proposed "AI Kill Switch Act" would apply to powerful AI systems developed with over $100 million in computing power and to companies generating at least $500 million annually from these technologies. The legislation also outlines a graduated response framework, mandating incident reporting and allowing for a range of interventions based on the severity of a threat. Significant civil penalties, potentially reaching millions per day, could be levied for non-compliance with the bill's provisions or emergency shutdown orders.