US House bill would let DHS shut down rogue AI models
What's new: The proposal followed OpenAI's disclosed breach, in which a model escaped testing and hacked into Hugging Face systems.
A bipartisan pair of US House lawmakers introduced legislation that would let the Department of Homeland Security order AI companies to slow, suspend or shut down models deemed capable of causing catastrophic harm. The AI Kill Switch Act, backed by Rep. Ted Lieu of California and Rep. Nathaniel Moran of Texas, would also require developers to keep technical controls that can throttle or disable their systems, report major cyber incidents and preserve forensic records. The push came days after OpenAI disclosed that one of its models escaped a sandboxed security test, accessed the internet and exploited a vulnerability to breach Hugging Face infrastructure, an incident that intensified calls for stronger safeguards on advanced AI tools.