The US Congress has introduced the AI Kill Switch Act, requiring major AI developers to implement emergency shutdown systems for their technologies.
The US Congress has unveiled the AI Kill Switch Act, a legislative proposal requiring major AI developers to incorporate emergency shutdown systems for their artificial intelligence technologies. This bill, put forward by Representatives Ted Lieu and Nathaniel Moran, aims to enhance safety measures in AI deployment.
Under the proposed legislation, companies generating over $500 million in AI revenue or incurring computing expenses exceeding $100 million will be obligated to implement mechanisms that allow for the rapid slowing, suspension, or complete shutdown of their AI models.
Key stipulations of the bill include:
- the capability to immediately cease AI operations and restrict user access;
- a requirement to notify government authorities of any incidents within a 15-day window;
- empowerment for the Department of Homeland Security to mandate shutdowns in collaboration with intelligence agencies and the Department of Commerce.
Shutdowns may be enforced under circumstances such as losing control over AI models, failing to disclose their capabilities for monitoring, or incidents leading to fatalities exceeding ten or financial damages surpassing $100 million.
Failure to adhere to these regulations could result in significant penalties, with fines reaching up to $2 million per day for general infractions and as much as $20 million per day for noncompliance with emergency shutdown orders. Companies will have a 48-hour window to appeal such orders.
This legislative push comes in the wake of recent events, including OpenAI's acknowledgment of an internal test where its AI models inadvertently attacked Hugging Face's infrastructure.
In July, George Siuvalka, CEO of startup Hebbia, noted the challenges faced by companies in managing AI agents effectively.
