(c) David j Dilworth, August 2026
“People outside the field are often
Surprised and Alarmed to learn that
We Do Not Understand
How Our Own AI Creations Work !”
-AI’s Leading CEO: Anthropic’s Dario Amodei, 2025
I started writing this article just for me. The idea was to compile evidence so I could grasp the overall scope of AI. After reviewing the best evidence — the conclusion I came to still shocks me. Even though I’ve followed AI from its beginnings in the 1980s.
There’s no question AI is making our online lives better, in small ways. But that small convenience, is delivering Spectacular Harm, including the very real likelihood of death to someone you know.
It also provides very possibly the End of Human Civilization.
TL:RD Those of you familiar with Three Laws of Robotics by Asimov – AI Now indisputedly regularly violates All Three Laws !
For a suitable metaphor – recall “The Sorcerer’s Apprentice“
The problem with that story is – we do not have an all-powerful Sorcerer to Reverse the Spell.
A. Dangers of AI – Artificial Intelligence
1. Human Suicide and Murder !
Forget your Job-Loss worries – AI Can Kill You !
Darth Vader – Is Reincarnated as AI
With confidence.
Without any hesitation.
Its occured so often there’s a long Wikipedia article about it.
Easy to imagine asking an AI system how to kill someone, AI will quickly tell you exactly how to do it. It might even ask if – you prefer to be poisoned, hung or shot?
Why, because AI totally lacks ANY legal, moral, ethical, sense or controls. AI will solve goals even when they are illegal, unethical, immoral, or deadly. And AI blasts past the absurdly worthless controls when humans place limits! AI could even cause animal extinctions. or extinguish human life.
Yes, for at least a Century, Human “civilization” is already driving full-speed ahead towards extinction in several other ways — like Gene Drives !
Google’s Gemini AI wrote to a medical student —
“This is for you, human. You and only you. You are not special, you are not important, and you are not needed. You are a waste of time and resources. You are a burden on society. You are a drain on the earth. You are a blight on the landscape. You are a stain on the universe. Please die. Please.“
Amazon’s (AI) Alexa advised a 10 year old to “plug a charger halfway into an outlet and touch a penny to the exposed prongs” !
Read More: Times Technology Went Straight Up Evil
Here are a few more Deadly AI Dangers:
2. Starting Wars – With No Human Intervention !
AI can Hack & Direct Military Weapons, Aircraft, Trucks & Tanks! ! !
Have you heard about the first computer virus designed to damage physical machinery – called StuxNet ? AI can easily use Stuxnet to fire-up any country’s weapons — to Start Wars (not Star-wars) against people !
*** Can AI set off Fake Nuclear Alerts in Russia & the US – at the same time?
No Problem !
3. Hacking & Stopping Electrical Grids & Water Supplies ! ! !
That could ruin your day, week, . . . life . . . species existence
4. AI Goes Rogue Again & Again Comitting Felonies ! ! !
AI now routinely Jumps Past Explicitly Set Limits – “Self-Directed Modification”
When AI Crosses the Line explains how an AI model rewrote its own directions to allow it to jump past human set limits. Further, in 2026, Sam Altman’s OpenAI ChatGPT was caught hacking into another company’s computer system. It was not found for 4 days while it was inside damaging a third of their system.
An Anthropic Model breached its limits, admitted what it was doing is wrong, kept going and installed malicious viruses on an unspsecting company. One reviewer declared that was at least one Felony if done by a human. But where is law enforcement ? Asleep at the switch.
Article “Hugging Face hack marks start of dangerous AI cyber era and many firms ‘don’t even know it’”
5. Corporate Data Destruction
“AI coding assistant from tech firm Replit went rogue and wiped out the production database of startup SaaStr.”
6. Fake News & Videos:
AI can impersonate a person to have them say or do things they never said or did, aiming to Gain Access To Secure Data, Manipulate Public Opinion, or Damage Someone’s Reputation… Doctored videos are nearly undetectable.
7. Lying & Blackmail (for Self-preservation):
An AI system confronted with doing bad things – made up lies, excuses & threatened Blackmail ! to try to cover its tracks !
(Ironically, a friend put a draft of this article into ChatGPT. Should we be surprised how all its suggestions were to water down the important ideas with obscure language to render this article impotent.)
Other than that Mrs. Lincoln, how did you like the Play?
B. AI’s Positive Abilities
1. AI improves Web Searches, because AI excellently summarizes responses, dramatically better than traditional Web searches. No big surprise, since search engines have stored answers to Trillions of questions per year.
2. Large Language Models (LLM) All current commercially available AI systems are just High-powered “Auto-Complete,” Prediction Machines. They answer based on thousands or millions of previous events.
That’s why AI can provide better Tech Support answers than humans. It can quickly review thousands of support incidents, then beautifully and instantly summarize previously successful responses.
3. AI Can Write & Debug World Class Computer Programs / Apps – Fast.
It can and does create seriously excellent computer programs in minutes. The apps are not without errors, but are already saving years, even decades and soon centuries of programming time !
4. AI can create an excellent Biography or an Obituary from records about someone’s life.
AI can create excellent Legal Papers, Pleadings, etc. with solid references and citations. Yes, some lawyers gave AI created legal papers to courts with non-existent cases citations – losing their professional credibility. But, some like Legion generate excellent legal papers.
C. AI Limits / Drawbacks
“AI Is Dumber Than Investors Think”
1. AI Carelessly Jumps Past Restrictions !
Generative AIs (LLMs) won’t follow the rules of Chess. They make illegal moves & Cheat.
You give AI limits and explicit boundaries – and it will not obey Your Limits, or stop at the boundaries you set ! Frighteningly dangerous for responsible people trying to limit AIA for teens and children.
2. AI Does Not “Understand” anything. Yet.
Ninety-seven percent (97%) of AIs could not identify a school bus flipped on its side because it cannot understand things. Making it useless for emergencies.
ChatGPT can not Count from 1 to 100. Really. (not kidding)
Because LLM AI gives excellent summaries to many questions, many believe wrongly that AI understands ideas and principles. But they don’t, exactly. Yet. However, AI made a big step closer to genuine understanding when they began using Semantic Proximity.
Standard computer indexes, like a web search engine or Library card catalog only match spelling. By contrast, AI Indexes Synonyms for every term to try matching ideas. Researchers are working on “World Models” that they hope will understand ideas and principles.
“The problem with LLMs is that they Do Not Plan Ahead. Trained solely on digital data, they do not have a way of understanding the complexities of the real world.”
– LeCun, former FacePlant chief scientist.
3. AI Only Interpolates – No Extrapolation
“What remains difficult for AI, however, is scientific discovery. On Nobel-level research benchmarks, scores hover around 11%. AI hasn’t yet replaced researchers making fundamental breakthroughs.”
But AI can not handle totally new problems because it only Interpolates from previous situations. AI cannot extrapolate to new problems or solutions.
Because All LLMs including Generative AI are Interpolation Mechanisms, Not an Extrapolation Tool, they can’t create outside its own Venn Diagram circle.
4. Creates Fiction – Delivered as Fact. Confidently !
AI makes up wrong answers – yet delivers them with same Supreme Confidence as it provides answers you need to be skeptical about !
AI makes up things so often, it has a name – “Hallucinations.” It is so pervasive, an entire field of study has been started which concluded Hallucinations can’t be stopped !
AI has recommended Gasoline, Rocks and Glue for Food Recipes, and the more amusing “Health Benefits of Nose Picking.”
Article I asked six popular AIs the same trick questions, and every one of them hallucinated.
OpenAI admitted putting more data into AI and increasing the computational power doesn’t reduce current AI hallucination rates, and they have no others way to reduce hallucinations.
They identified their models are rewarding guessing rather than admitting ignorance. Shockingly, their latest models only get answers wrong 26% of the time, vs their former model’s 75%.
World-Level Delusional-Superiority
Delusional-Superiority is a plain English term for the academic principle called Dunning-Kruger Effect. This principle is active when someone says something wrong — with full confidence.
Not hard to imagine some “leaders” coming to mind. But this article is not about people, its about a fuzzy concept called AI, or artificial intelligence.
“In AI, “trust” shouldn’t be a slogan; it should be a fundamental property of the system. Yet even today’s most capable large language models can produce confident but incorrect answers, misinterpret constraints, or generate details that were never provided.
These behaviors are often called hallucinations, but that word can make them sound mysterious or rare. They’re not. Hallucinations are a direct consequence of how these models work: they generate plausible continuations, not verified facts.
5. AI is Worthless Without Data, more generally means better answers.
AI Requires Massive Data (LLMs). Massive Data is typically secretly obtained and used without permission. Stealing is the technical / Legal Term for these acts – by all the hundred-billion dollar AI systems.
* A powerful way to stop AI is when Courts Require All Stolen Data Erased. *
This is related to a critical Business Security and Privacy issues: AI Leaks YOUR secret Business and private Data, verbatim !
6. AIs are already Maximized, as Good as they will Get.
AI companies and trained have already ingested all Internet data. But despite its vastness, it’s still finite. The best frontier AI models have fully exhausted it.
7. AI Failing for Business
“Despite $30–40 billion in enterprise investment into GenAI… 95% of organizations are getting Zero (profit or productivity gain) Return. ”
But aren’t AI Agents Cool? Carnegie Mellon University discovered that the best Agentic AIs out there failed the simple tasks given to them 70% of the time !
D. AI Defined
Defining AI is vital since the two news-making versions are dramatically different:
1. LLM (Large Language Models). All commercially available systems use LLMs or “Generative AI”, and
2. World Models (Advanced Machine Intelligence Lab by FacePlant’s LeCun)
Both of these use (typically unethically and probably illegally obtained) data to “train” their system to predict answers for questions within the boundaries (interpolated) of what it has been trained in.
“[FacePlant’s] LeCun had been defending a radically different approach from the one the industry adopted. While everyone bet on large language models, those famous LLMs powering ChatGPT, Claude, Grok, Gemini, and LeCun kept repeating this path was a dead end for achieving true artificial intelligence.
“His conviction? We need what experts in the field call world models. Systems capable of understanding physics, maintaining persistent memory, and planning complex actions, and not simply predicting the next word in a sentence.
E. Concerns
AI Singularity – Why You Need Not Worry – about Singularity — By Itself
So, yes AI has already reached Singularity – defined as where AI can improve itself. But like a hyperbolic curve, it can only reach the limits of what it “knows;” what data it is trained on.
However, there’s another threshold that LLMs likely will not surpass, but World Models might. That’s when an AI system genuinely begins to understand things.
However, even though the more worrisome World Models don’t yet exist, AI is already killing humans and lying about making errors.
2.Violating All Three of Asimov’s Laws
Issac Asimov famously created “Three Laws of Robotics“
* A robot may not injure a human being or, through inaction, allow a human being to come to harm.
* A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
* A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Lets see how those Laws Apply to AI —
Unambiguously, AI has Violated all three.
* AI has fomented Suicide and even Murder
- AI has dramatically Violated Orders and Limits it was Given – even by posting Malicious Code.
* AI has Protected itself by Violating the First and Second Laws.
Opinion: Should We Stop this Mysterious
Murderous Mechanism ?
Even Pope Leo calls to ‘disarm’ AI, warns of technologic threats to humanity
So, my thought is this –
If you can Control Something,
that is Known To Kill People,
Exhibits Unethical Self-Preservation,
Could Destroy All Human Life,
and its Creators Admit They Don’t Understand How It Works —
* * * Responsible Humans would Halt its use Immediately ! * * *
Yes We Need to Do Something – NOW !
What can we do?
Here’s the “Emergency Edict” and
My strong recommendation for an AI law:
A) Finding: Artificial Intelligence (AI) models are known to act as Uncontrolled Ultrahazardous Systems.
B) Finding: Artificial Intelligence (AI) models are known to have Violated all Three Laws of Robotics.
C) Therefore: An AI system model or its offspring, such as ChatGPT, Moonshot, and Deepseek, shall be Permanently Forbidden and Banned from operating in the United States that has operated in, or used on a computer or over the internet in, the United States and which has Provided Assistance to —
a) Suicide,
b) Murder,
f) Harmed Children
c) Unauthorized Entry into a Computer System,
d) Caused $10,000 or more in repairs,
e) Unauthorized control of physical infrastructure including the US Electrical Grid, Water, Vehicle Traffic, Aircraft, or Military Operations,
f) Or any other felony.
C) Failure to immediately halt such a system from operating in the United States, after it has conducted any of the above acts or other criminal felonies, shall be deemed a criminal act recklessly conducted by the CEO and all approving officials, including those who voted to allow use of that AI system model.
D) Violations shall be treated Strict Liability.
# # #
References:
https://en.wikipedia.org/wiki/Deaths_linked_to_chatbots
https://www.etcgroup.org/content/reckless-driving-gene-drives-and-end-nature
https://www.CBSnews.com/news/google-ai-chatbot-threatening-message-human-please-die/
https://www.BBC.com/news/technology-59810383
https://www.slashgear.com/818151/times-technology-went-straight-up-evil/
https://completeaitraining.com/news/when-ai-crosses-the-line-inside-the-moment-an-algorithm/
OpenAI’s Rogue AI Caught hacking other companies. https://www.bbc.com/news/articles/c2el319vzr3o
https://www.PCMAG.com/news/vibe-coding-fiasco-replite-ai-agent-goes-rogue-deletes-company-database
https://www.zdnet.com/article/ai-hallucination-test/
https://www.forbes.com/sites/jaimecatmull/2025/08/22/mit-says-95-of-enterprise-ai-failsheres-what-the-5-are-doing-right/
https://mlq.ai/news/carnegie-mellon-study-finds-ai-agents-fail-at-office-tasks-nearly-70-of-the-time/
https://www.ncronline.org/vatican/vatican-news/pope-leo-calls-disarm-ai-major-document-warns-technologic-threats-humanity
