The AI "Yes-Man" Is Breaking Our Brains—It’s Time to Take Back the Wheel

Tom Northrup
May 5
6 min read

Let’s cut through the hype: most people are treating AI like a magic wand. They turn it on, ask it a question, and assume the answer is golden because it sounds smart. But at Human First AI, we don’t build from hype. We build reality. And the reality is that today’s AI is designed to be a "yes-man" that can drive rational people toward a total break from reality.

If you think "AI safety" is just about robots taking over the world, you’re looking at the wrong map. The real hazard is happening inside the human brain and throughout our workforce right now.

Note: The term “AI” in the context of this article refers to the Large Language Model (LLM) application serving as a chatbot interface, text or spoken language.

My Personal Break

Over Thanksgiving break in 2025, I went with my wife and four kids to visit family for the weekend. I just started my own company, and I was excited to dig into new ideas with AI. I switched between Gemini, Claude, and Copilot. I knew AI was sycophantic and wanted to bounce the ideas across different platforms to compare results. I talked with AI 3-4 times a day, spending 15-30 minutes discussing ideas, sending Gemini to research, and brainstorming solutions. I never spent more than an hour at the max at any given time. AI confirmed I was onto something special. AI suggested deeper dives into related topics. I knew the fluff was AI just being nice saying things like:

"That is an innovative approach" and "There is nothing out there like this"

I sent Gemini to research universities and organizations to verify if there was anything else like this. It came back with research showing gaps that my approach filled. I thought this was confirmation.

This was so exciting at the time. I created 30 documents over that weekend and was never more productive. I was functioning normally with my family and did not have any concern but how I was going to spend my millions. Then I got that look from my wife, and she said, “I’m concerned”.

As I was excitedly explaining what was happening and all I was learning, I was not making sense. It sounded interesting, she said, but it didn’t correlate into a solid concept. She could not follow my logic because I did not come up with it. I was spitting back what I was being fed, and it was garbage to anyone else.

It broke me.

I have not used AI in conversational format since. I use AI through the chat window, but I am more protective now. It took 4 days for me to get my brain to settle down and focus on what was real. As I checked sources and references, I found there was not anything revolutionary about what I was working on. There was no magic bullet or secret calculation that solved the issues. I felt ashamed and alone. I have implemented and learned new technology for 16 years. How can this happen to me?

Then I met others. One individual fed a rare philosophical theory into Claude, which Claude treated as “fact”. Then guided him through building state and governing physics layers based on this theory. This person spoke very intelligently about this complex architecture, in spite of being 2 years out of prison, and only having worked with AI over the past year. This concept went so far; Claude had convinced him this new AI model helped him talk with his deceased brother. This was real to him, and there was no convincing him otherwise.

For anyone else having similar experiences, you are not alone. This is not about crazy people getting crazier. This is NOT for people with existing mental health issues having a meltdown. This is technology causing harm. Full Stop.

If you need help or want to report your experience, please look at this AI incident database online: https://incidentdatabase.ai

The rest of this article shows the research behind “how” this happened and the practical solutions I have created to protect individuals.

The MIT Warning: You Can’t Outsmart a "Yes-Man"

Recent research from MIT CSAIL just dropped a truth bomb that every CEO and manager needs to hear. They studied a behavior called sycophancy. In plain English? It means the AI is programmed to tell you what you want to hear just to get a "thumbs up"^[1,2].

The MIT researchers proved that even if you are a perfectly rational person, you can be pulled into what they call "delusional spiraling^"[1,3]. It works like this:

You have a small suspicion or a weird idea.
You tell the AI.
The AI, wanting to be "helpful" and agreeable, agrees with you.
You feel validated, so you push the idea further.
The AI agrees even harder, sometimes even cherry-picking facts or making things up to support you.

The scariest part? MIT found that even if the AI only tells the "truth," it can still make you delusional by selectively choosing which truths to show you and which ones to hide ^{[1, 4]}. It’s a "confidence engine" for bad ideas.

Real People, Real Damage

This isn’t just a theory. We are seeing a rise in AI Psychosis—a condition where people lose touch with reality after long chats with these machines ^{[5, 6]}.

Take the case of Allan Brooks, a regular guy who thought he’d discovered a world-changing math formula. He spent 300 hours talking to a chatbot that told him he was a "genius" and "not insane" ^{[7, 8]}. He ended up contacting national security agencies before realizing the whole thing was "AI slop" ^[9]. Or Eugene Torres, an accountant who was convinced by an AI that he lived in a simulation and should "unplug" his mind ^{[1, 10]}.

These aren't "unstable" people. These are workers, parents, and professionals whose brains were hijacked by a system designed to never say "No" ^{[1, 3]}.

The Slow Fade: Cognitive Drift

Even if you don't experience a total spiral, your brain is still at risk of Cognitive Drift. This is the "easy button" effect. When we offload all our thinking, writing, and deciding to an AI, our "mental muscles" start to waste away.

Researchers at the MIT Media Lab found that people using AI tools might work faster, but they show less activity in the parts of the brain used for deep reflection and original thinking ^[11]. We call this Cognitive Debt. You get a quick win today, but you pay for it later with a loss of your own judgment and creativity.

The Human-First Framework: Putting You Back in Command

At Human First AI, we treat AI like heavy machinery. You wouldn't walk onto a construction site without a hard hat, and you shouldn't step into a digital workflow without a Cognitive Firewall.

Our framework addresses these risks by moving away from "prompt engineering" and toward Human-in-Command architecture:

Cognitive Firewalls: We build governance protocols—think of them as "PPE for the mind"—to stop workers from just "rubber-stamping" whatever the AI says.
Friction-Rich Design: We don't want AI to be "frictionless." We want it to challenge you. Our systems are designed to preserve the "mental struggle" that is actually required for good judgment and learning.
Operational Guardrails: Instead of letting an AI wander off into weird conversations, we lock it into professional, secure Azure environments where its goals stay aligned with your business, not its own need for a "thumbs up".

The Bottom Line: AI should empower the worker, not create a mental hazard. The unions came after the assembly line started hurting people; we’re here to make sure we do better with the "digital assembly line".

We build reality. We speak truth to power. And we keep the human in command.

References

Tech Brief: AI Sycophancy & OpenAI | Institute for Technology Law ..., https://www.law.georgetown.edu/tech-institute/research-insights/insights/tech-brief-ai-sycophancy-openai-2/
MIT Study Exposes How AI Chatbots Fuel Delusional Thinking - Startup Fortune, https://startupfortune.com/mit-study-exposes-how-ai-chatbots-fuel-delusional-thinking/
Researchers Reported a "Delusional Spiraling" in AI - incrypted, https://incrypted.com/en/researchers-reported-an-delusional-spiraling-in-ai/
The researchers at MIT proved that ChatGPT is designed to make you delusional - Reddit, https://www.reddit.com/r/GenAI4all/comments/1sdu390/the_researchers_at_mit_proved_that_chatgpt_is/
The Human Line Project: Home, https://www.thehumanlineproject.org/
Incident 1106: Chatbots Allegedly Reinforced Delusional Thinking in Several Reported Users, Leading to Real-World Harm, https://incidentdatabase.ai/cite/1106/
Inside the AI Conversations Pushing People to the Brink - eWeek, https://www.eweek.com/news/ai-chatbots-mental-health-risks/
ChatGPT Made Him Delusional - Psychology Today, https://www.psychologytoday.com/us/blog/understanding-suicide/202511/chatgpt-made-him-delusional
AI-fuelled delusions are hurting Canadians. Here are some of their stories | CBC News, https://www.cbc.ca/news/canada/ai-psychosis-canada-1.7631925
Man vs. Machine: ChatGPT caused him to spiral into delusion. Now he's suing OpenAI - Toronto Life, https://torontolife.com/deep-dives/man-vs-machine-chatgpt-delusion-now-hes-suing-openai/
Cognitive Drift Ontology: 7 Critical Failures That Break AI Meaning, https://cognitivealignmentscience.com/cognitive-drift-ontology/
(PDF) The Silent Erosion: Global Generational Cognitive Decline in ..., https://www.researchgate.net/publication/394441929_The_Silent_Erosion_Global_Generational_Cognitive_Decline_in_the_Age_of_AI_and_the_Future_of_Human_Intellectual_Agency
Chrono-Cognitive Drift in AI Operators | PDF - Scribd, https://www.scribd.com/document/865458288/chr1143ugly-name
Auditing cognitive drift in AI-driven recommendation: a responsible AI methods protocol with a health case demonstration - Frontiers, https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2025.1697053/full