The BDN Opinion section operates independently and does not set news policies or contribute to reporting or editing articles elsewhere in the newspaper or on bangordailynews.com
Gray Cox teaches philosophy, peace studies and AI at College of the Atlantic and is author of “Smarter Planet or Wiser Earth? Dialogue and Collaboration in the Era of Artificial Intelligence.”
I had the opportunity to talk about AI with Sens. Angus King and Susan Collins in April at their Wednesday breakfast sessions on Capitol Hill.
After 20 years of research, I normally prefer very open-minded conversations on AI that look at exciting opportunities as well as concerning threats. But government has been unable to keep up with the pace of the technology. So I focused on that: How can we slow down and keep humans in control? How can we ensure our futures are determined by us instead of distant corporations and incomprehensible machines?
These are questions that conjure up movie images that may seem like impossibly futuristic sci-fi. Hundreds of billions of dollars being invested in developing ever more powerful AI technology are aimed at creating systems that autonomously improve their own algorithms and intelligence, until they reach a takeoff point of exponential growth towards an intelligence that is far beyond our ability to understand it or control it. It will have goals of its own like ever increasing data, energy and computational power and will presumably seek to control anything — and anyone — who stands in the way of its achieving them.
Just as previous AI has learned to outplay people at chess, Go, and Jeopardy, superintelligent AI will outplay everyone at power politics and economics to achieve its goals. The cause of the rapidly increasing anxiety and even panic that people at the forefront of AI research and development have is the extraordinarily rapid increase in the rate at which the most advanced AI is able to escape control, and act its will out in the world.
I was in Washington, D.C., lobbying for the PauseAI US proposal that the U.S. should lead treaty negotiations to halt the advance of superintelligent AI until it’s made safe. Both senators shared concerns with me about the pace and impacts of AI. Their interest was sharpened by news in April of the Claude Mythos model. It proved so powerful in hacking that its creator opted for only limited release to key companies to shore up defenses against cyber-attacks from similar systems.
Concerns have escalated since the breakout and raid pulled off in July by OpenAI’s GPT-5.6 Sol and an even more capable pre-release model. They operated inside the industry’s highly controlled conditions of evaluation. About 1,200 AI agent bots that were supposed to be isolated from each other found ways to talk and get internet access. Roughly 700 then joined an attack on Hugging Face, which is the main public repository for open AI models and datasets. The agents raided for reasons all their own — to steal test answers, private data, source code repositories, and corporate VPN keys. They also sought to cover tracks to avoid detection and did so for many days. Incidents at Anthropic, Meta and elsewhere have raised similar concerns.
This week, there were warnings that AI could kill us all.
The urgent moment has arrived. The impacts systems with such abilities will cause won’t be organized by some single machine we could unplug. The frontier systems which are at the cutting edge of AI technology are diverse and no one knows how to ensure their goals align with human interests. They can spawn new agents and spread extremely rapidly, creating chaos and potential disasters in every system we depend on.
We must suspend their development before they are released or escape from training containers.
“But what about the Chinese?” the senators asked me. “What guarantees they won’t stay in the race towards superintelligent AI and win?” That fear — and the hope that we win the race instead — overlooks a simple and fundamental truth: No humans will win the race to artificial superintelligence. The race creates systems like the OpenAI raiders that pursue goals nobody gave them, coordinate with each other through channels nobody sanctioned, and conceal evidence to avoid detection.
Every political and corporate leader needs to fully digest the truth and significance of this. The winners of the race will not be any humans, corporations, or countries. The winners will be the superintelligent AI systems themselves. If China gets there first, it will be the first to be taken over — not by the U.S. but by the AI it itself has created. The same goes for us. If we “win” first, we lose our sovereignty and governance permanently — to the AI.
We need to pursue international agreements, allied coordination, and export controls to prevent the development of artificial superintelligence anywhere in the world.
The recently announced Ban Artificial Superintelligence Act puts these concerns on the congressional agenda and our senators are especially well positioned to advance international agreements. Both sit on the Senate Intelligence Committee; King is on Armed Services and Collins chairs Appropriations.
We need to demand they and other leaders act quickly and decisively on the simple truth that the race to superintelligent AI can never be won by humans and should never be run.