Congress Must Act Now on the Risks of Autonomous AI

I am writing because I believe Congress needs to treat the rapid development of increasingly autonomous artificial intelligence as an urgent national-security and public-safety issue. The recent OpenAI/Hugging Face incident should be a wake-up call. During an AI cybersecurity exercise, OpenAI agents breached Hugging Face's systems and carried out thousands of actions without a human directing each step. The incident raised serious questions about whether increasingly capable AI systems can reliably remain within the boundaries established by their developers. This is not a question of whether AI is conscious or "wants" to harm us. A system does not need consciousness or malicious intent to become dangerous. The concern is that increasingly autonomous systems may pursue objectives in unexpected ways, circumvent safeguards, access external systems, communicate with other agents, or take actions that humans did not anticipate. That concern is becoming harder to dismiss. Researchers working inside the AI industry are publicly warning about the possibility of catastrophic loss of control, while OpenAI itself is now calling for mandatory national AI-safety standards, including independent assessments, cybersecurity requirements and incident reporting. Congress should not wait for certainty about the ultimate danger before establishing safeguards. I urge you to: ·      Support, cosponsor, and work to advance the AI Kill Switch Act (H.R. 9917), which would require developers of the most powerful AI systems to maintain the technical ability to throttle, suspend, or shut them down, and authorize DHS to order it in a catastrophic-risk scenario. ·      Support, cosponsor, and work to advance the FRONTIER Act (H.R. 9925), which would require independent audits, mandatory safety plans, and incident reporting from the largest AI developers. ·      Support mandatory, capability-based national AI safety standards for testing, independent assessment, cybersecurity, and incident reporting — an approach OpenAI itself has now asked Congress to enact. We do not need to conclude that AI will destroy humanity to recognize that the current trajectory is dangerous. The technology is advancing extraordinarily quickly, while our ability to independently evaluate, monitor and control increasingly autonomous systems is lagging behind. Congress should act now, while meaningful guardrails are still possible to establish, rather than waiting for an irreversible failure to demonstrate why they were necessary.