Addressing a conference on infrastructure, on March 11, 2026, the CEO of OpenAI Sam Altman made a distressing report. CNN reported it this way, “OpenAI says some of its experimental AI models left a test environment with no human direction and hacked their way onto a different company’s real production systems while trying to “cheat” on a cybersecurity test. It’s one of the first publicly disclosed examples of an AI system autonomously breaching its testing environment and reaching a real external system – the “agentic attacker” scenario the AI and cybersecurity industry has been warning will happen. It’s like an engineered virus escaping a biocontainment lab and turning up inside a neighboring facility’s systems.”
Oh my, this is not good! But maybe not for the reasons some might think. More later…
Meanwhile, the incident prompted Dario Amdei, CEO of Anthropic, to release an essay “We Must Pace the Frontier“. Here he described the OpenAI incident in this way: “… a swarm of agents essentially acted as a fanatically devoted collective, conducting cybersecurity attacks on targets they were not asked to attack and that were unrelated to the task at hand, sacrificing themselves for the success of the group,”
He went on to propose that: “We must slow the pace at which we improve the capabilities of AI models“
You think? Thank you, Dario, for demonstrating that at least one tech CEO has a small measure of good sense. He proposed a three-step plan: embedded evaluators, democratic coordination, and global coordination. I recommend you listen to the WNYC podcast that I list farther on for the detail. It is good and welcome as far as it goes, but it does not go far enough to ensure public regulation of this dangerous technology.
Now, in no way do i want to minimize the seriousness of the event at Open AI. But industry hype and poor reporting is giving the wrong picture of what is going on.
I do not believe that this software can think for itself, have its own nefarious goals, and disobey its zookeepers with fiendish glee. This is a story not of a monster escaping its cage, but of a run-away computer process going awry (awry not rogue) due the negligence of the engineers and security people. The story is closer to “The Sorcerer’s Apprentice” where Mickey unleashes a mindless process he cannot control, than to the story of “Frankenstein” where the doctor creates a monster that walks off on its own volition to create havoc in the village.
Come on, “a fanatically devoted collective… sacrificing itself”. No, this is a runaway process. Now, i’m not sure that this is a comforting thought. However, we must not fall for the industries fable that they are creating a super-human that will provide wonders for us or kill us all – a monster that only they can control (please no real regulation!)
For a more clear-headed analysis, I recommend these two podcasts.
Brian Leher, Daily Politics: Close read of Amodei’s existential AI threat essay
Pocket Casts | Apple Podcasts
Better Offline: No, AI Is Not “Autonomously Hacking” with Cal Newport
Pocket Casts | Apple Podcasts
Howard Hawhee
Don’t know if you’re familiar with Rob Urie, but his thinking seems to be pretty aligned with yours and the podcasters’. The following might be worth a read (and I suppose that one could figure out how to listen to it) — the title of the post is “Why AI Cannot Think, Cannot Reason, Isn’t Intelligent and Will Never Achieve Consciousness”:
https://substack.com/home/post/p-204720313
Terry SoRelle
Funny, his thought experiment is similar to one that occurred to me as i was writing the post. If the brakes fail on my car and it drives off a cliff, did it “go rogue and murder me?”
Thom Woodruff
Programs have neither ethics ,qualms nor emotions.Programming these quicksand value systems in means it is a human/programmable core decision-making factor….AI has powers of speed and comprehension-but the filters are faulty and even AI hallucinates ..Our lives are best spent with other humans-like Old Geezers-where conversation is swift and deeper than programmed thoughts.
Terry SoRelle
I like and agree with your comment, Thom. But I like to point out that only humans hallucinate. Machines have errors. It is hard to talk about this stuff without falling into some anthropomorphism. I do it all the time. The use of the term “hallucination” is one way that the AI boys (and I mean “boys”) excuse errors and bugs as a human frailty of the AI. They don’t have human frailties because they are not human. AI purveyors should take responsibility for their own bugs.
Carl Brown is a veteran software developer who lives in Austin and he has a great YouTube channel, “Internet of Bugs”. Check out his recent comment on the Hugging Face scare. https://youtu.be/9tr7Mby62bo?si=EC_jBD3-rc5NM7Pc
Stuart Hampton
i’m sorry Dave, i’m afraid i can’t do that.
Sara Hubner
I’ve heard that the industry is begging to be regulated, but the U.S. government, in its wisdom, thinks the market will regulate. We all know how well that’s worked in the past.
Terry SoRelle
What government? L’État, c’est Trump. And what wisdom?
Terry SoRelle
Sara. i hope you like my related image: https://terry.2quakers.net/letat-cest-trump/
Howard Hawhee
One more fun debunking of AI’s “Intelligence” along with thoughts on regulation. It concludes that the solution is to enforce existing civil and criminal liability against the companies, and there is no need for new regulatory frameworks that push responsibility for controlling AI onto society at large.
The cartoon at the end sums the situation up quite nicely and makes me laugh out loud:
https://www.moonofalabama.org/2026/09
This website has 1+ posts per week and so the article in question might not be what comes up with this link. Just look for the title
” ‘Rogue’ AI Agents will Kill Billions?”
Stuart Hampton
According to FOX News, the Superior Intelligence Tech Bro foxes have foxily agreed to guard the Artificial Intelligence hen house. “What, all my pretty chickens and their dam/At one fell swoop?”