Letter to the Editor: The danger of AI is real

Dear Editor,

I read with interest your Nov. 18 article on Senator Mark Kelly’s new framework to protect Americans from the most advanced AI models.

Senator Kelly is right about the problem. Companies should not grade their own homework. Self-regulation has failed, and oversight that only arrives after an incident is not oversight.

But his proposed fix still misses the core issue. It governs the model, not the action.

His framework calls for a rulebook of hazard levels, independent examiners embedded at AI companies to review safeguards as models are built, and testing and liability before and after deployment. None of that stops a consequential action at the moment it happens. A test can pass, an examiner can sign off, and a company can disclose a risk, and the AI system can still make the transfer, delete the file, or send the data.

The danger is not only how smart a model is. It is what authority it has been given to act. A safe model with write access is dangerous. A dangerous model with no authority does nothing.

That is why my team and I are building Pulpo here in Arizona. Governance has to sit before execution, as a narrow, one-use permit for that one specific action, with durable evidence afterward. No permit, no governed effect. Anything else is monitoring dressed up as prevention.

I respect Senator Kelly’s intent and his service. But if we want to protect Americans, we need to govern what AI is allowed to do, not just how it is tested or watched.

Sincerely,
Mike Diegel
Pima, Arizona