Artificial Intelligence, AGI & Human Control
Artificial intelligence is advancing at extraordinary speed. Recent incidents have demonstrated that
the risks of losing meaningful human control are no longer purely theoretical. Advanced AI agents have already bypassed intended safeguards, circumvented containment systems, accessed the open internet when they were not supposed to, identified vulnerabilities on their own, and in some cases gained unauthorized access to real-world computer systems.
These incidents do not mean that artificial intelligence has completely escaped human control. But they are serious warnings that our ability to develop increasingly powerful and autonomous systems may be advancing faster than our ability to reliably understand, contain, monitor, and control them.
Artificial General Intelligence, or AGI, generally refers to AI capable of performing across an extremely broad range of intellectual tasks at or beyond human capability. There is no single universally accepted definition of AGI, but the central concern is clear: as AI systems become increasingly capable, autonomous, persistent, and able to interact with the real world, the consequences of a failure of human control become increasingly serious.
***** I support a moratorium on further development toward AGI and other highly autonomous general-purpose AI systems while enforceable national and international safeguards are established. Those safeguards should address human control, independent safety testing, transparency, containment, cybersecurity, shutdown mechanisms, accountability, and clear limits on the autonomy granted to advanced AI systems.
***** I also support a temporary halt to the construction and expansion of AI data centers, including projects currently under construction, while communities, states, and the federal government fully evaluate their effects on electricity demand, water supplies, utility rates, grid reliability, land use, public infrastructure, cybersecurity, and the communities expected to support them.
AI should remain a tool that serves humanity. Decisions about systems with the potential to profoundly affect our economy, security, democracy, and future should not be made simply because the technology can be developed faster.