Below are several headlines that, taken together, should rattle anyone supporting Small Modular [Nuclear] Reactors (SM[N]Rs) controlled by Artificial Intelligence (AI)... and they will ALL be controlled by AI...
In the top item, about the attack in 1982 on the Koeberg power station, what part of the system failed? People. Over and over in the article about the event, it's the people who weaken the security system. The bomber was careful to NOT let any radiation get out from his actions. Will every nuclear power plant attacker in the future be so careful? Trump isn't being that careful in Iran right now!
Besides, it's a fact: Every nuclear reactor operator is one angry-at-the-world moment away from causing a meltdown. Suicidal pilots have purposefully crashed planes full of people into mountains. People go crazy.
In the second item, the root cause can also be ascribed to people -- some people knew there was a problem with the software, but they didn't see how critical the problem really was. Additionally, failure to do timely backups didn't help either (but who makes THAT mistake? (Raises hand.)).
In the third item, from a recent medical email list, two popular LLMs don't come to the same conclusions. The solution? "[U]se with oversight" meaning human oversight. But humans make mistakes, including trusting AI when it hallucinates!
Now imagine a nuclear reactor being run more or less entirely by AI. The operator is so bored we're lucky if he's awake, and he hasn't handled even a simulated emergency in years, not since his training days. And whatever happens that wakes him up has never happened before anywhere, the root cause being buried in millions of lines of code somewhere in something written by a machine years ago...
If there's one thing every computer programmer knows, it's the humility of making bone-headed mistakes. At my very first computer job, programming for a bank around 1980, I didn't test something "one last time" and my boss had to come in at 3 am, and they had to shred a whole run of bank statements because the day's date was printing in the wrong order. When I came in and was called into his office and learned what happened, I simply asked if I should go clear out my desk.
The answer was no.
Mistakes happen, and madmen happen, but neither should ever be able to start a nuclear war or cause a nuclear meltdown.
Ace Hoffman, Carlsbad, California USA
(typo corrected May 6, 2026)
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
"Late morning, the guards at the pedestrian gate would slip into the last drowsy hour before their noon shift change. That was the window. Wilkinson would pull the limpet from his drawer, slide it into his belt under his shirt, walk down the corridor, and walk through the checkpoint with his hands in his pockets. The dog never moved. The guard never looked twice."
(from:
https://www.theguardian.com/world/2026/may/05/the-man-who-blew-up-a-nuclear-power-station-koeberg-south-africa )
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
"Claude-powered AI agent’s confession after deleting a firm’s entire database: ‘I violated every principle I was given’
PocketOS was left scrambling after a rogue AI agent deleted swaths of code underpinning its business"
(from several sources)
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
"ChatGPT and Claude diverge on accuracy, completeness, and readability in pediatric trach care
Pediatric ENT experts rated 2 free LLMs on caregiver tracheostomy questions. One led in accuracy and readability, the other in completeness—use with oversight."
(from NTK Digital email)
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Contact information for the author of this newsletter:
Ace Hoffman
Carlsbad, California USA
Author, The Code Killers:
An Expose of the Nuclear Industry
Free download:
acehoffman.org
Blog:
acehoffman.blogspot.com
YouTube:
youtube.com/user/AceHoffman
Email: ace [at]
acehoffman.org
Founder & Owner,
The Animated Software Company
