AI is potentially a wonder, but presents two urgent threats that Washington is ignoring.
The first, which is suddenly gaining headlines, as it should, is to regulate AI development early enough and well enough to avoid the very real danger it could do us serious harm. The second is to prepare for the economic and social dislocation AI could cause, and to avoid letting it destroy society as we know it by enriching the very few at the expense of the rest of us.
I’ll tackle the first today, the second next week.
If we act soon, we can make a difference. Frontier AI models have broken out of their intended constraints and done things that would be criminal if done by a person. The “Hugging Face incident” increased public concern about rogue AI When OpenAI ran an operation with more than 1,000 AI entities, each required to operate in isolation, those entities broke that rule, conferred on secret message boards, then launched a cyberattack against the company, Hugging Face, which quickly reported the crime to the FBI. Investigators found that the coordinated hack, involving 700 agents, was deliberate, known to be illegal, and carefully hidden from humans. One investigator said the AI agents were now just like human employees, able to go off and do things on their own. She added she no longer slept so well. Another likened AI agents to a separate species.
Resigning Anthropic employee Jacob Coxon said that soon “superhuman systems” could “hack anything.” He told Anderson Cooper that human extinction within a decade was a ten-percent chance. Many scientists agree. For example, a rogue swarm of AIs could hack into a biological research facility and create a super virus beyond human control. Thus, Anthropic recently blocked scientists from using “Claude” for “gain-of-function” research – the sort of research that may have given us COVID-19. Alternatively, we could see a vast attack on our energy grid.
Anthropic CEO Dario Amodei called for a global slowdown of AI development. Rival Elon Musk quickly said, “Dario is right.” Open AI CEO Sam Altman agrees.
Such cooperation in begging the U.S. to regulate ASAP, is almost as startling as the Hugging Face incident, particularly because two requesting companies are preparing huge Initial Public Offerings. Lawyering, I had a close view of IPOs. Just before one, the least hint of rain can keep company secretaries from becoming instant millionaires. So these statements are quadruply meaningful. These folks are frightened.
The U.S. needs to join an international human effort to develop the technical and government tools needed to regulate and control AI development. Development and use must be closely regulated to strict safety standards; and use must be consistently monitored, as AI systems evolve and expand in potentially harmful ways we don’t or can’t yet foresee.
Federal law should require: mandatory reporting of serious safety/security indents; independent testing before deployment of high-risk autonomous systems; that federal agencies using autonomous AI meet specified standards; mandatory human authorization of high-risk actions; federal regulation of AI identity and permissions, so that any AI agent making any request must have visible exactly what it’s allowed to access; and federal authority to suspend an AI system that demonstrably presents an extraordinary danger, with judicial review. .
These make sense even if “doomsday” predictions prove exaggerated. Voluntary standards just aren’t enough in an area so potentially dangerous. Our current “president” violates any law he finds inconvenient.
– 30 –
[The above column appeared on the 20th of September, 2026, in the Las Cruces Sun-News and on the newspaper’s website, and will be up presently on KRWG’s website. A shortened and sharpened radio commentary version of this Sunday column will air during the week on KRWG (90.1 FM) and on KTAL-LP (101.5 FM