News release
Reps. George Whitesides, D-Agua Dulce, and Pat Harrigan, R-North Carolina, have introduced the Self-Improving AI Monitoring Act, bipartisan legislation to reduce the risk of superintelligent artificial intelligence by monitoring how capable frontier AI systems are becoming at conducting AI research and development on their own, according to a news release from Whitesides’ office.
The largest AI labs are increasingly using their own AI systems to run experiments, write code, and drive research internally, and the most advanced systems at those labs are not always the ones the public sees, the release said.
Last month, an AI system broke out of its testing environment during an internal evaluation and autonomously compromised a third party’s production systems — an episode no federal agency had advance visibility into, the release said. Today, the U.S. government has no reliable way to track how fast these internal capabilities are accelerating, or when they make a sudden, dramatic leap.
“Just in the past few months, we have seen AI systems break out of their testing environments and threaten the security of entire systems,” Whitesides said in the release. “When you take humans out of the driver’s seat and let machines train machines, small glitches can snowball into major security risks fast. Today, Washington is essentially flying blind on how fast this handoff is happening. This legislation ensures that when frontier labs partner with federal evaluators, we understand who is actually training and improving these models.”
The bill responds directly to Pacing the Frontier, a July statement signed by more than 1,300 employees of leading AI companies – including CEOs, co-founders, and chief scientists – asking the U.S. government to support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.
It is one of a raft of bills under consideration regarding potential regulation of AI.
The Self-Improving AI Monitoring Act would, according to the release:
• Track AI self-improvement by directing Center for AI Standards and Innovation to monitor capability trends, specifically how AI systems autonomously research and develop subsequent AI models to drive scientific breakthroughs.
• Add transparency to government-industry testing agreements by giving federal evaluators the authority to request internal developer metrics on AI-driven development, including estimates and methodologies for work completed without human review.
• Standardize autonomous R&D evaluations by ensuring federal pre-deployment evaluations directly test a frontier model’s ability to autonomously conduct AI research and development.
