The Honorable Mark R. Warner The Honorable Brian E. Schatz United States Senate Washington, DC 20510 c/o Elizabeth Falcone 703 Hart Senate Office Building, Washington, DC 20510 c/o Reema Dodin 722 Hart Senate Office Building, Washington, DC 20510 cc: The Honorable Barack Obama Dear Senators Warner and Schatz, This pertains to AI regulation and slowdown. I have one of the longest records of AI & Society academic work in the US, currently at CWRU in Cleveland; I have the longest record on AI for a scholar coming out of Hawaii; I have been associated with NSA and DARPA AI projects, Harvard, Stanford, and recently LAW OF AI, adding to the EU AI Liability proposal. I have also done DoD research for every major US combat branch and a considerable amount of computer security. The risks to the nation with AI escaping its sandbox, crossing air-gaps, or just accessing the internet like a cabal of super-hackers are real: a) infrastructure is fragile and concentrated; b) cyber defenses are inherently porous; c) taking control and doing pernicious deeds will be easy. This is a national security and cyber security issue, not just an AI technology and commerce issue. This framing should help achieve legislative weight despite international competition and GDP concerns. AI will commit cyber crime on a massive scale at zero-day speeds. The superintelligence word is hype. Killing all humans might be possible but quite improbable. AGI is not serious. Those are clickbait. Still, it won't take much to do damage to essential systems. International adversaries see opportunities without needing to be particularly smart. This was General Nakasone's fear and it is now happening for real. The cybersecurity solution is to protect those systems from AI inputs the same way we protect them from foreign domains: a) Require layered firewalls to sandbox AI models and agents effectively; b) Filter RAG internet access to short, simple non-write payloads; c) Restrict, delay, and audit agent creation and agent-to-agent communication by law; d) Close the responsibility and intent loopholes in the CFAA. Do you remember when the first internet worms arrived and smart people had to go to jail for releasing them? "The intent to release an uncontrollable agent into a shared ecosystem is inherently criminal" -- GEMINI. This makes the political action easy. Push bipartisan interest in security. The larger problem is social system stability. This requires more work and more time on many fronts. I write this to you because you two are the most important to me. My brother in Hawaii worked with your father, Dr. Irwin Schatz; I know his Tuskegee work from Detroit in 1965. My nephew in Indianapolis received Senator Warner's biography when the kid started golfing with Senator Todd Young's family. Nice lawyers, but I wanted him to know Mark Warner's path out of the Midwest: the best Senator in my lifetime in my opinion (and I've known Obama since 1972). Your staffers can file this opinion among the many who are similarly concerned. Mine comes from exactly 51 years of AI research and philosophy, starting with spell-checking at Punahou School, where Senator Schatz got his start too. This opinion is informed by AI Seminars with our elite minds at CWRU, and our Provost-level AI Committee. We seek truth, not hype, and many on this campus are concerned. Sincerely, Ronald P. Loui, PhD Adjunct Full Professor, Computer and Data Sciences, CWRU, Cleveland, OH