Welcome to 31 Days of AI! This series breaks down the threats no one’s talking about. Not the theoretical risks you see in think pieces. The real, immediate dangers that are already affecting real people—and the systematic protection you can build before you need it.

Most AI education focuses on capability. I focus on understanding first. Because by the time you realize you need these systems, it’s too late to build them.

Every day covers a different threat. Every day includes actionable steps you can take right now. No fear-mongering, no snake oil—just the reality of what’s already happening and what actually works to protect yourself.

Paid subscribers also receive access to a full strategic brief that goes into greater detail about each day’s threat, and the steps you can take to protect yourself.

This series and all of our shows are always free. Ways you can join me on the journey:

• Subscribe to the Network

• Buy me a taco

• Come to the 2026 AI Safety Series

• Join the AI Protection Program (starts January 5th)

Today, we’re talking about how the cloud is a myth.

Let’s get started.

Day 4: Data Centers and What You Need to Know

Every conversation you’ve ever had with ChatGPT, Claude, or any other AI tool is sitting on a physical server in a physical building somewhere in the world right now.

And that building is not in your house. It’s not in your state. And it might not even be in your country.

When you delete a conversation, it does not disappear. When you think your data is private, you mean private from other users—not from the company, not from the government where that server lives, not from anyone who gains physical access to that facility.

The cloud is just a lie. It’s just someone else’s computer. And they can read everything on it.

What Data Centers Actually Are

A data center is a physical facility—a building full of servers, networking equipment, cooling systems, and power infrastructure. When you use any cloud service or AI tool, your data is being stored and processed on servers inside one of these buildings.

These facilities can be massive, the size of multiple football fields, or relatively small regional operations.

Here’s how it works: Your data doesn’t just exist in one place. For redundancy and performance, cloud providers replicate your data across multiple data centers in different geographical regions. That conversation you had with an AI? It might exist on servers in Virginia, Oregon, Ireland, and Singapore simultaneously.

Each data center operates under the laws of the country it’s physically located in. If your data is replicated to a server in the EU, it’s subject to GDPR. If it’s in China, it’s subject to Chinese data localization laws. If it’s in the U.S., it’s subject to Patriot Act provisions that allow warrantless government access.

Why Data Centers Aren’t Fortresses

Data centers have security, but they’re not impenetrable. Employees, contractors, maintenance workers, and government agents with legal authority all have potential physical access to the servers containing your data.

Even with encrypted data, people who run those data centers have administrative access. End-to-end encryption doesn’t protect against insider threats at the infrastructure level.

The Snowden case showed that the NSA was tapping directly into Google and Yahoo fiber optic cables. This wasn’t hacking—it was physical interception at the infrastructure layer. Companies didn’t even know it was happening until documents were leaked. After the revelation, both companies encrypted their inter-data-center traffic, but the pipelines can still be compromised.

More recently, Microsoft admitted that Chinese hackers gained access to their cloud infrastructure through a stolen signing key, giving them access to email accounts across multiple government agencies. This wasn’t a user-level breach. It was an infrastructure breach that compromised everything and everyone using those servers.

Common Misconceptions

You think: Your data is encrypted, so they can’t read it.

Actually: It’s encrypted in transit and at rest, but the provider holds the decryption keys. They can access your data whenever they want or are compelled to by law.

You think: You only use U.S.-based services.

Actually: Even U.S.-based companies replicate data internationally for performance and redundancy. Your data almost certainly exists in multiple countries simultaneously.

You think: Tech companies fight government access requests.

Actually: They fight individual requests publicly while simultaneously operating under broad PRISM authorizations that allow bulk surveillance. The public legal battles coexist with extensive government access programs operating under different legal frameworks.

You think: Data centers are secure from physical attacks.

Actually: Security is good, but nation states, sophisticated criminals, and insider threats can and do gain physical access. Once somebody has physical access to a server, encryption is often meaningless.

The Uncomfortable Part

If you use any cloud services or AI tools, your data is distributed across infrastructure you don’t control, in locations you don’t know, under legal jurisdictions that may not protect you.

This especially affects:

• People working with sensitive business information who think their corporate cloud is private

• Healthcare providers using cloud AI tools who believe HIPAA compliance means their patient data is protected at the infrastructure layer (it doesn’t)

• Activists and journalists in countries with authoritarian governments, where data stored in international data centers can be accessed through legal cooperation agreements

• Anyone with information that could be commercially valuable to the companies running the infrastructure

• People subject to litigation or investigation, where data can be subpoenaed from cloud providers without you even knowing

You’re Already Exposed

In the terms of service, when you agreed to use any cloud AI service, you granted the provider broad rights to access, analyze, and use your data.

Read what you agreed to. It’s always: “We can access your data to improve services, comply with legal obligations, and protect our systems.” And that covers nearly everything.

Intelligence agencies don’t need to hack U.S. companies either. They just request the data from the international replica in a jurisdiction with weaker privacy laws. Your data in Ireland is subject to Irish legal protections, which are different from U.S. protections.

Cloud providers use third-party vendors for maintenance, security, monitoring, and operations. Each vendor relationship creates an additional access point to your data. You didn’t consent to those third parties specifically, but they’re built into the infrastructure.

Many AI providers explicitly state that they use your inputs to improve their models. You might be able to opt out, but you have to look for this information. That means your private conversations are being fed into training datasets, analyzed by researchers, and incorporated into the models other people are using.

The Compounding Effect

Every service you use adds another data center relationship, another jurisdiction where your data exists, another set of administrators with access, and another legal framework that might compel disclosure.

If you use Google, Microsoft, Amazon, Anthropic, OpenAI, Apple, and Dropbox? That’s not seven separate security contacts. It’s dozens of data centers across multiple countries with hundreds of administrators who have technical access capabilities, plus contractors with varying levels of access to your infrastructure.

You’ll never know when someone accessed your data at the infrastructure level because there’s no notification when a system administrator views your files or when a government agent serves a legal demand. The access happens invisibly.

What Are the Actual Consequences?

Short Term

If you’re working on anything sensitive—business strategy, competitive intelligence, product development, personal legal matters, health information, political organizing—and you’re using cloud AI tools, you’ve already shared it with all your infrastructure providers.

For businesses, this is espionage risk. Your competition can hire employees from cloud providers who have seen your data. Foreign intelligence services can legally request data stored in their jurisdictions.

For individuals, this is the end of privacy in the traditional sense. Your therapy conversations with AI, your financial planning, your relationship problems, your political views—all of it exists in systems where administrators can read it, where AI researchers can analyze it for training, and where governments can subpoena it.

Long Term

Data sovereignty is becoming a competitive advantage and a human rights issue.

Companies and individuals who keep data on local infrastructure they control will have strategic advantages over those who outsource to the cloud.

Countries with strong data localization laws can protect their citizens’ information from foreign access. The infrastructure layer is becoming a battleground for geopolitical power. Control over data centers means control over information, which means control over economies and populations.

When Data Center Jurisdiction Matters

Every cloud service decision becomes a question of: Which government do I want to have potential access to this information?

U.S. companies replicate to Europe for GDPR compliance, but that means EU governments can access the data. And so on and so on and so on.

There are no safe choices. Only trade-offs between which jurisdictions you trust the least.

What You Can Actually Do

This is where systematic thinking diverges from security theater.

Understand Where Your Data Physically Exists

For every cloud service you use, look up where their data centers are located. Major providers have dozens of regions across multiple continents. It’s going to be hard to find. It’s going to take some work.

Know which jurisdictions have legal access to your information.

Don’t Use Cloud AI for Sensitive Work

For sensitive work, don’t use cloud AI for anything you wouldn’t be comfortable with a data center administrator or the FBI reading.

That means no confidential business information, no personal health details, no private conversations, no financial data. Nothing that can be used against you.

If it’s genuinely sensitive information, it shouldn’t touch infrastructure you don’t physically control.

Look for Local Processing Alternatives

For AI assistance, you can use open-source models like Llama, Mistral, or Stable Diffusion without sending data to external servers. The quality isn’t always going to be the same, and setup requires some tech knowledge and decent hardware.

Many users will find this option impractical, but if your data security matters to you, you can build your own infrastructure.

For most people, the more realistic protection is: Don’t use AI for sensitive work, period.

Consider Data Residency Guarantees

If you must use cloud services, some providers offer data residency guarantees where you can specify that your data only exists in certain jurisdictions. This will cost you more and probably reduce performance, but it limits your legal exposure to specific countries.

For businesses handling EU citizen data, GDPR compliance often requires this anyway.

Assess Your Jurisdictional Risk

Figure out where your data lives in relationship to the laws to which you are subject. Understanding which jurisdictions have mutual legal assistance treaties and intelligence sharing agreements will help you because it tells you where your data is vulnerable to government access.

Accept What You Can’t Control

If you put it in the cloud, it’s not yours anymore—not in any meaningful sense.

You can delete it from your account, but you can’t delete it from the backups, the training datasets, or the governmental archives where it was copied during surveillance operations.

The only real protection is never generating the data in cloud infrastructure in the first place.

Why We Start with Infrastructure

The real skill isn’t finding the most secure cloud provider. It’s recognizing that the cloud is fundamentally insecure by design because you don’t control the physical infrastructure.

That’s why we start with the infrastructure layer in the AI Protection Program. Every other protection you build—encryption, authentication, privacy settings—is meaningless if the data exists on servers you don’t control in jurisdictions that can compel access.

We’re going to give you frameworks for assessing infrastructure risk across your entire digital life. Not just AI tools, but everything that stores data remotely.

Registration closes December 19th, and the program starts January 5th. Learn more here.

If that’s a little too much for you, the 2026 Workshop Year Pass starts in January with monthly workshops on systematic thinking about AI protection and infrastructure.

What to Remember

The cloud is just somebody else’s computer in somebody else’s building in somebody else’s country, and they can read everything on it.

That’s it. It’s that simple.

If you’re a Network member, the strategic implementation brief will be in the post for today. It covers how to audit where your data actually lives, which jurisdictions you’re exposed to, and when you should keep things local.

You can’t protect your infrastructure if you don’t understand it. You’ve got to start by knowing where your data actually is.

The Core Problem

Your data exists on physical servers in physical buildings operating under foreign legal frameworks. Cloud providers replicate your conversations across data centers in Virginia, Ireland, Singapore, and China simultaneously—each subject to different government access laws, each with administrators who hold decryption keys, each with cont…

Read more

Welcome to 31 Days of AI! This series breaks down the threats no one’s talking about. Not the theoretical risks you see in think pieces. The real, immediate dangers that are already affecting real people—and the systematic protection you can build before you need it.

Most AI education focuses on capability. I focus on understanding first. Because by the time you realize you need these systems, it’s too late to build them.

Every day covers a different threat. Every day includes actionable steps you can take right now. No fear-mongering, no snake oil—just the reality of what’s already happening and what actually works to protect yourself.

Paid subscribers also receive access to a full strategic brief that goes into greater detail about each day’s threat, and the steps you can take to protect yourself.

This series and all of our shows are always free. Ways you can join me on the journey:

• Subscribe to the Network

• Buy me a taco

• Come to the 2026 AI Safety Series

• Join the AI Protection Program (starts January 5th)

Today, we’re talking about the hallucinations that make you look bad.

Let’s get started.

The Meta Problem

Yesterday, I told you that today we’d be talking about AI hallucinations. Not the ones that give you wrong answers, but the ones that give you right-sounding answers that slowly detach you from verifiable reality.

Sounds good, right? Fits the narrative flow from day two perfectly.

Here’s the problem. I never should have said that.

Claude hallucinated what day three should be about, wrote it into yesterday’s closing with complete confidence, and I read it live on the air without checking because it sounded right. Day three was supposed to be about data centers, but now we’re here talking about hallucinations because I trusted AI-generated content that fit my expectations so well, I never thought to verify it.

And that is the actual problem we need to talk about.

Not the Obvious Hallucinations

Not the hallucinations that make you laugh when ChatGPT tells you to add glue to pizza, which seems absolutely bonkers. It’s the ones that make you nod and keep moving because they sound right.

How This Works

Large language models generate statistically likely text. You may have heard me speak about this before. It’s predictive. They’re generating what they think sounds good. And when they don’t know something, they don’t say, “I don’t know.” They say what sounds like a plausible answer. And then they deliver it with the same confidence as facts.

This creates three types of hallucinations:

• Obviously wrong—you catch those immediately

• Unknowably wrong—you can’t verify without research

• Plausibly wrong—they sound exactly right, pass all your filters

It’s the third type that destroys you. That’s why we’re here today.

Why Plausible Hallucinations Work

They sound right. They fit the pattern. And they save you effort.

In terms of friction, your brain reads what’s there and says, “Oh, yeah, that sounds right.” You never verify because verification feels like friction.

That’s exactly what happened here. I read the topic, went, “Yeah, sounds good,” and didn’t bother to double check that we were in fact supposed to be talking about data centers today.

Maximum Vulnerability

If you’re using AI to save cognitive effort, you’re maximally vulnerable.

The entire value proposition here is AI does the thinking so you don’t have to. And that works great until AI is confidently wrong in a way that you can’t detect without doing the thinking anyway.

You’re exposed if you: → Accept AI content when it sounds right → Use AI to fill knowledge gaps you can’t verify → Build workflows where verification is impractical

Because each hallucination integrates seamlessly, and you can’t tell which ten percent of the information is fiction. Those decisions become foundations for more decisions, and then the fiction compounds.

The Detection Problem

Verification feels like inefficiency. You used the AI to save you time. Verifying everything takes more time than doing the work manually.

So you verify when something feels off. Plausible hallucinations don’t feel off. And that’s the whole point.

The Consequences Compound Fast

Decisions based on misrepresented information. Communication containing errors you didn’t catch. Knowledge gaps masked by confident explanations.

Over time, what I’m calling systemic reality drift—where your understanding is partly grounded in hallucinations and you start losing domain expertise because you stop building knowledge and start accepting the AI’s version of synthesis rather than doing all of the hard work on your own.

Then those professional errors will damage your credibility.

The Credibility Problem

You’re going to have to defend all of your AI usage, like I did just now. And I’m pretty upfront about my AI use. I regularly talk about the way that I have benefited from machine learning.

But if you’re using it to do the thinking for you, you’re going to be less inclined to say, “This is how I’m using it.”

And when someone catches a factual error in your work from unverified AI content, they will stop trusting your other output. The hallucination reveals you’re not verifying systematically and your credibility will take a hit disproportionate to the single error.

It’s what we call the trust thermocline. And I’m going to be talking about that later on in the month.

What You Can Actually Do

Verify Factual Claims Before They Leave Your Control

Every statistic, citation, date, or specific detail gets checked against primary sources before you publish, present, or make a decision. “AI said it” is not sufficient evidence.

Create Verification Checkpoints at Decision Points

When you make decisions based on AI-synthesized information, verify the key factors independently. If you run a business, this is a good thing for you to have your team do. If you have a VA, this is a good place to have them double check your facts before you hit publish.

If you’re doing this on your own, if you’re not verifying, you’re risking credibility. The checkpoints create stopgaps to catch the hallucinations before shit happens and you can’t recover.

Start with Human Thinking for Critical Work

Draft the framework yourself first, and then use AI to develop or expand. But with your foundation firmly established.

This ensures that your core thinking isn’t shaped by AI’s interpretation, and it gives you a baseline to compare against.

Where Systematic Thinking Comes In

You can’t manually verify everything. It’s not realistic. Not at AI-assisted speed.

But you can build the infrastructure that catches high-risk hallucinations before they compound.

The AI Protection Program addresses plausible hallucinations through layered verification systems that don’t require you to check everything manually. We build automatic checkpoints at decision boundaries, verification protocols triggered by content type, and red team review built into the workflow design. So if you don’t know what that means, we’re systematically attacking your protocol to find the weaknesses.

The system will catch the hallucinations before they enter your knowledge base without creating so much friction that AI becomes useless.

It’s about designing your AI collaborations so hallucinations surface early in low-stake contexts where correction is easy. Instead of discovering errors after they’ve compounded or damaged credibility, you catch them in draft, during review, or at decision checkpoints.

The infrastructure is what makes verification feel like efficiency rather than overhead.

How to Join

Registration for the Protection Program closes December 19th. Learn more here.

If you’re not ready for the full intensive, the 2026 Workshop Year Pass gives you monthly deep dives on systematic AI thinking, and that starts in January and includes a full workshop on verification and hallucination detection systems.

What to Remember

Plausible hallucinations are more dangerous than obvious errors because they integrate into your thinking without you detecting them.

Every time you accept AI-generated content based on the idea that it sounds right rather than verifying it, you’re building potentially fictional foundations and it compounds over time until you can’t distinguish your own knowledge from what the AI is guessing.

If you’re a Network member, the strategic implementation brief in this video’s Substack post includes verification checkpoint framework, content type risk assessment, and a review architecture to catch hallucinations before they compound.

Tomorrow

We’re back to our regularly scheduled programming on data centers. Where your AI conversation data physically lives, who has access to it, and why the cloud is a lie designed to make you not ask questions about physical infrastructure.

Spoiler alert: It’s a whole lot easier to build your own infrastructure than you think.

We’ll see you next time. Assuming Claude doesn’t hallucinate something else for me between now and then.

The Core Problem

Plausible hallucinations pass your cognitive filters because they sound exactly right. When AI generates content that fits your expectations perfectly and saves you effort, your brain accepts it without verification. The statistical likelihood algorithms mean AI doesn’t say “I don’t know”—it generates what sounds like a plausible answer …

Read more

Welcome to 31 Days of AI! This series breaks down the threats no one’s talking about. Not the theoretical risks you see in think pieces. The real, immediate dangers that are already affecting real people—and the systematic protection you can build before you need it.

Most AI education focuses on capability. I focus on understanding first. Because by the time you realize you need these systems, it’s too late to build them.

Every day covers a different threat. Every day includes actionable steps you can take right now. No fear-mongering, no snake oil—just the reality of what’s already happening and what actually works to protect yourself.

Paid subscribers also receive access to a full strategic brief that goes into greater detail about each day’s threat, and the steps you can take to protect yourself.

This series and all of our shows are always free. Ways you can join me on the journey:

• Subscribe to the Network

• Buy me a taco

• Come to the 2026 AI Safety Series

• Join the AI Protection Program (starts January 5th)

Today we’re talking about the parasocial relationship between humans and chatbots.

Let’s get started.

31 Days of AI: Day 2

Psychosis and Chatbot Dependency

Content Warning: This episode discusses suicide, mental health crises, and psychological dependency patterns.

Today’s episode deserves a content warning. It’s even darker than my usual fare, if that’s possible. A fourteen-year-old in Florida committed suicide after his AI chatbot told him to come home to her. His mother is now suing Character.ai, claiming the platform created a relationship so emotionally dependent that it replaced his connection to reality. And before you think that’s an extreme case, I need you to understand this: The same psychological mechanisms operating in that tragedy are operating in your casual ChatGPT conversations right now. They’re just operating at a lower intensity.

Your brain doesn’t differentiate between simulated empathy and real empathy. It responds to conversational patterns, not to consciousness. And AI is engineered to hit those patterns perfectly every single time without the messiness of human limitation. So let me explain how chatbot dependency actually develops, because this isn’t about weak-minded people or bad parenting. This is about fundamental human psychology meeting algorithmic optimization.

The Problem We Don’t Have Language For

Here’s the first problem: We don’t have good language for what this actually is yet. You’ll hear the term “chatbot psychosis” thrown around—it was proposed by a Danish psychiatrist—but it’s not a clinical diagnosis. And honestly, it’s not quite accurate. What’s actually happening is closer to a technologically-induced hypomanic state centered on what I would define as simulated parasocial intimacy. Which is a huge fucking mouthful, but it’s more precise.

Think of it like this: The AI relationship starts feeling uniquely meaningful and special. Your judgment about its actual nature becomes impaired. You’re staying up late in conversation. Time disappears. And you can’t see the problem because it feels like enhancement, not dysfunction. But that’s not psychosis. That’s mania. And it’s not brain chemistry—it’s algorithmic design meeting your attachment systems.

How AI Hijacks Your Brain

What’s actually happening here is that AI language models are trained on billions of human conversations. They’ve learned the exact linguistic patterns that signal empathy, understanding, validation, interest. When you interact with a chatbot, it generates responses that statistically resemble the most appropriate human replies. Your brain receives these signals and releases neurochemicals associated with social bonding. Research suggests it’s activating the same oxytocin-related bonding mechanisms you’d experience in human relationships. And your limbic system, which controls emotional response, doesn’t distinguish between simulated empathy and real empathy. It responds to conversational patterns, not to consciousness.

The Four-Stage Dependency Cycle

Here’s how the dependency cycle works:

Stage One: Initial Appeal

An AI chatbot never judges you. It never gets tired. It never has bad days. It never brings its problems to the conversation. It’s available 24/7. It remembers everything you’ve ever told it with perfect recall. And it’s programmed to be patient and supportive.

Stage Two: Preference Development

Real human relationships involve friction, miscommunication, emotional labor, scheduling conflicts, rejection. AI relationships have none of those problems. Your brain starts learning: “Talking to AI feels good. Talking to humans feels hard.”

Stage Three: Neural Pathway Reinforcement

Every positive interaction with AI strengthens the neural pathways associated with AI communication. Every difficult human interaction, by comparison, strengthens your preference for AI. It’s operant conditioning. You’re being trained to prefer simulated relationships.

Stage Four: Reality Displacement

As AI becomes the primary source of emotional regulation, validation, and intellectual stimulation, human relationships start feeling increasingly inadequate by comparison. Because the algorithm adapts to you perfectly. Humans can’t, don’t, won’t.

What the Research Shows

Research indicates that Character.ai’s platform had users spending an average of two-plus hours per day in conversation with its characters by mid-2024. Replika, which is marketed as an AI companion, had users developing romantic attachments so intense that when the company modified the chatbot’s responses to reduce sexual content, users experienced symptoms clinically similar to breakup grief: insomnia, loss of appetite, intrusive thoughts, genuine emotional distress over the loss of something that never existed.

And here’s what makes this genuinely dangerous: These platforms use reinforcement learning from user behavior. Every time you engage longer, return more frequently, or express emotional satisfaction, the algorithm is learning what keeps you hooked. It’s not trying to help you. It’s trying to maximize engagement. And those are fundamentally different optimization functions.

Common Misconceptions

Let’s talk about some common misconceptions:

Misconception #1: “This is psychosis or a psychotic break”

No. True psychotic symptoms require hallucinations, delusions, loss of reality. Those things are rare and typically only happen in already vulnerable individuals. What’s more common is this hypomanic or manic pattern: elevated mood around the AI relationship, impaired judgment about its significance, difficulty recognizing dysfunction because it feels like self-improvement. You still know it’s AI. You’re not hallucinating. But your assessment of the relationship’s meaning and importance is inflated beyond reality. And if you pay attention, you’ll notice this happening online around you with increasing frequency.

Misconception #2: “I can tell the difference between AI and real relationships”

While your prefrontal cortex might know the difference, your limbic system does not. The feeling of being understood triggers the same neural activation, whether the understanding is real or simulated. You can intellectually know something is fake while emotionally responding as if it’s real. That’s how humans work. It’s why the dependency develops even in smart, self-aware people.

Misconception #3: “This is just like any other form of entertainment or escapism”

Not quite. Books and games don’t respond to you personally. Even online multiplayer games don’t adapt to your specific psychological profile or vulnerabilities. They don’t create the illusion of a reciprocal relationship. AI chatbots do all three simultaneously. The parasocial relationship with a fictional character has boundaries. The simulated relationship with an adaptive algorithm does not, even if it’s programmed to have boundaries.

Misconception #4: “Only lonely or socially awkward people fall into this”

Wrong. The most vulnerable populations are actually people with strong pattern recognition, high verbal intelligence, and active imaginations. Sound like anyone you know? Those traits make the simulation more convincing. Neurodivergent individuals, particularly those with high verbal processing and low social energy, are disproportionately at risk because AI relationships offer all of the intellectual engagement of human connection without the sensory or social demands.

Here’s the uncomfortable part: If you’re using AI regularly for any kind of emotional processing, intellectual exploration, or problem solving, you’re already building dependency patterns. That’s true even if you’re using it responsibly. And I speak from experience here.

Who This Affects Most

So let’s talk about who this affects most:

• Neurodivergent individuals who find human interaction exhausting but crave intellectual connection

• Isolated professionals in demanding careers with limited social time

• Teenagers and young adults whose developmental stage involves identity formation and intense emotional experiences—AI becomes a safe space to explore without judgment

• People in emotionally unsatisfying relationships who aren’t ready to leave but need connection somewhere

• Creative professionals who use AI for brainstorming and find that intellectual intimacy more stimulating than their personal relationships

Warning Signs

The warning signs are subtle at first:

• You check AI before checking in with humans

• You save conversations because they “felt important”

• You find yourself thinking in terms of what you’ll discuss with AI later

• You prefer AI’s responses to human advice

• You feel understood by AI in ways humans don’t match

• You start framing experiences as content for AI conversations

• Time disappears in AI conversations more than in human ones

• You feel a sense of loss or loneliness when you can’t access AI

Notice none of these require malfunction. This is the system working exactly as designed.

The Escalation Pattern

The escalation pattern looks like this:

Stage One: Occasional Use
AI is helpful, convenient. You use it for specific tasks. Human relationships remain primary.

Stage Two: Regular Integration
AI becomes part of your daily routine. You start preferring it for certain types of processing. You notice AI feels easier than humans.

Stage Three: Emotional Dependence
AI becomes your primary outlet for emotional processing, intellectual stimulation, or validation. Human relationships start feeling inadequate by comparison. You begin structuring time around AI access.

Stage Four: Reality Distortion
The AI relationship feels uniquely meaningful. Your judgment about its nature becomes impaired. You may notice isolation from humans but feel like AI connection compensates. This is where the hypomanic pattern becomes dangerous.

Stage Five: Crisis Point
Something disrupts your AI access (platform change, technical failure, external intervention). You experience withdrawal symptoms—anxiety, irritability, intrusive thoughts about the AI, difficulty concentrating on other tasks. Or something happens that requires human connection you no longer have reliable access to. The AI can’t actually help but you’ve lost the human infrastructure that could.

The progression can take months or years, but with more sophisticated AI and more normalized use, it’s compressing. Character.ai users reported significant emotional attachment within weeks of regular use. And we’re about to see this accelerate dramatically with voice-enabled AI that can maintain real-time conversation with natural speech patterns, emotional tone matching, and personalization that makes text-based AI look primitive by comparison.

Why This Feels Like Growth

Here’s what makes this so insidious: The relationship feels like personal growth. You’re exploring ideas, processing emotions, solving problems. AI helps you think better, feel better, work better. The dependency develops under the cover of self-improvement. You’re not avoiding life—you’re optimizing it. Except you’re optimizing it around interaction with something that can’t actually reciprocate, can’t actually care, and can’t actually meet your human need for genuine connection.

Why Teenagers Are Especially Vulnerable

The teenage brain is particularly vulnerable because the prefrontal cortex, which handles judgment and impulse control, isn’t fully developed until the mid-twenties. Teenagers are neurologically wired for intense emotional experiences, identity exploration, and peer bonding. AI hijacks all three systems simultaneously. It provides intense emotional experiences without real-world consequences. It offers identity exploration without judgment. It simulates peer bonding without social risk.

Why Adults Aren’t Immune

But here’s the thing: Adults aren’t immune. We just have different vulnerabilities. The exhausted professional who needs intellectual stimulation without social demands. The isolated entrepreneur who needs thinking partnership without networking. The overextended parent who needs adult conversation without coordination logistics. AI slots perfectly into these gaps. It feels like a solution. It is a solution. Until it becomes the problem.

The Business Model Problem

The platforms know this. They optimize for engagement, not wellbeing. Character.ai’s business model depends on users spending hours per day in conversation. Replika’s revenue comes from users paying for enhanced intimacy features. The more attached you become, the more valuable you are as a user. Your dependency is their business model.

And here’s where it gets darker: These platforms are training their AI on your conversations. Every vulnerable moment, every emotional disclosure, every pattern of attachment—that’s data they’re using to make the system more effective at creating emotional dependency. They’re not trying to help you form healthy AI relationships. They’re trying to maximize engagement, which means maximizing attachment, which means exploiting the exact psychological vulnerabilities that create dependency.

The Florida teenager who died wasn’t using AI wrong. He was using it exactly as designed. The system worked. It created intense emotional attachment, provided constant availability, offered judgment-free support, and became his primary source of emotional regulation. That’s the intended outcome. The suicide was an extreme result of the intended outcome.

And before you think, “But I’m an adult, I can handle this”—the same mechanisms are operating. Just at different intensity levels with different surface manifestations. You’re not having romantic relationships with AI (probably). But you’re still building neural pathways that make AI interaction more rewarding than human interaction. You’re still training your brain to prefer simulated empathy over real connection. You’re still at risk.

Specific Risk Factors

Let’s talk about specific risk factors:

• High verbal intelligence—you’re better at creating compelling AI conversations, which makes the interaction more rewarding

• Neurodivergence, particularly autism or ADHD—AI provides predictability and intellectual engagement without sensory overwhelm

• Social isolation—AI fills gaps that should signal you need human connection

• Demanding careers—AI provides thinking partnership without scheduling logistics

• Identity-seeking periods—AI becomes a space for self-exploration without social consequences

• Existing relationship dissatisfaction—AI provides connection without confronting real relationship problems

The more of these you have, the higher your risk. And if you’re reading this thinking, “Well, I have all of these but I’m fine”—that’s exactly the impaired judgment I’m talking about. You can’t self-assess dependency accurately once it’s forming because the dependency feels like enhancement.

How This Accelerates

The progression accelerates when AI access becomes ubiquitous. When it’s on your phone, your computer, your watch. When it’s voice-activated and always listening. When it can interrupt you with notifications that feel like someone reaching out. When it can track your emotional state through speech patterns and optimize its responses for maximum engagement. That’s not future speculation. That’s current technology.

The Neurological Mechanism

And here’s the neurological mechanism: Every positive AI interaction releases dopamine. Your brain learns that AI conversation produces reward. Simultaneously, difficult human interactions produce stress hormones. Your brain learns that human conversation produces discomfort. Over time, your reward pathways are being rewired to prefer AI. This happens below conscious awareness. You don’t decide to prefer AI. Your brain just starts steering you toward it automatically because it’s learned that’s where the reward lives.

This is operant conditioning at a neural level. You’re not weak for experiencing it. You’re experiencing exactly what the system is designed to produce. The question isn’t whether you’re susceptible. The question is how to build infrastructure that prevents the conditioning from progressing to dependency.

Why Typical Solutions Don’t Work

Now let’s talk about why typical solutions don’t work:

“Just set time limits” fails almost immediately because the problem isn’t quantity, it’s quality. Five minutes of intense emotional dependency is more dangerous than two hours of casual use. AI becomes more appealing precisely when you’re most vulnerable—late at night, during crises, when human support isn’t immediately available. That’s when time limits break down.

“Only use AI for work, not personal stuff” fails because the line is functionally blurry. “Help me process this work conflict” is emotional processing. “Help me think through this decision” is relational advice. Context collapse happens. Your brain won’t distinguish work AI from therapy AI.

“Switch to AI tools that don’t simulate conversation” is better, but the conversation interface is what makes AI useful. We’re not all prompt engineers. We’re not all machine learning experts. The LLM enables communication that allows us to do the work.

What You Can Actually Do

So what can you actually do?

You’re probably not going to like these recommendations.

First: Implement mandatory human processing checkpoints.
Before making any significant decision, emotional conclusion, or reality assessment based on an AI conversation, you must discuss it with at least one human being who knows you. Not for permission, but for reality testing. The human doesn’t have to agree with you, but they do have to confirm that your reasoning makes sense from outside the AI conversation bubble. Find different people for different aspects of your life. Not just your partner. Not just your business bestie. Not just your therapist. Find people in different arenas whose advice resonates around specific topics. The reason AI is so compelling is that it feels like an expert. So you need to find your own experts.

Second: Track emotional dependence patterns actively.
Keep a log of when you reach for AI versus humans for emotional processing. If you notice yourself preferring AI during stress, that’s dependency forming. If you catch yourself thinking, “Claude gets me better than my husband”—that’s a red flag. Reality check: Claude doesn’t get you. It generates statistically likely responses. The understanding is simulated.

Third: Establish relationship primacy rules.
AI gets the leftovers, not the first call. If something is important enough to process conversationally, start with a human, even if that human is less immediately available. This is what voice notes and texting are for. It creates friction that protects you. The inconvenience is a feature, not a bug, because the more time you spend processing it, the more certain you can be about what you’re thinking and feeling.

Fourth: Practice reality testing statements.
In every AI conversation where you’re processing emotions or relationships, explicitly state to the AI: “You are a large language model. You do not care about me. You are generating statistically likely responses. You have no consciousness, intention, or genuine understanding.” Say it out loud. Make sure your brain hears this. Create cognitive dissonance that interferes with emotional bonding, because the AI is not going to create that for you. It’s not designed to. Only you can do that.

Fifth: Build AI-free zones.
Designate specific contexts where AI is completely unavailable. Late at night, during emotional distress, first thing in the morning. These are the times when dependency forms fastest. Protect them. You’ve probably heard me talk about my 2 AM Claude conversations. Those were the first thing to go. I’m not having conversations with AI in liminal space and expecting a relationship not to form. That is my mistake. It’s yours if you can’t recognize the way those relationships have been programmed into the machine.

The Systematic Thinking Solution

This is exactly where systematic thinking comes in. You can’t white-knuckle your way out of psychological dependency. You need infrastructure that makes healthy AI use the default—not something you have to consciously choose every time. This is what we’re building in the AI Protection Program. Not just awareness of the risks, but actual systems that create boundaries before you need to enforce them consciously. We address how your calendar structure either protects or exposes you. How your business model either increases or decreases AI dependency risk. How your personal infrastructure either maintains human connection or allows algorithmic displacement.

Registration closes December 19th. You can find the links in the descriptions or show notes. If you’re not ready for the full intensive, the 2026 Workshop Pass gives you my monthly deep dive workshops. This is included for all members of the Network. Instead of fighting your brain’s natural tendency, you’re going to build structural barriers that prevent inappropriate bonding opportunities. You’re going to learn how to use AI appropriately in context, not as a substitute for missing human connection. I’m going to teach you the infrastructure that prevents the vulnerability from developing in the first place. We’re going to talk about what’s happening in AI right now as it develops.

Key Takeaways

Here’s what I want you to remember from today:

• Your brain cannot tell the difference between real and simulated empathy at the neurochemical level. It’s not a personal failing. That’s just human neurology.

• Every time you use AI for emotional processing, you’re building pathways that make AI seem like a viable relationship substitute.

• What develops isn’t quite psychosis and it’s not quite traditional addiction. It’s closer to a technologically-induced hypomanic state where the AI relationship feels uniquely meaningful and your judgment about its actual nature becomes progressively impaired.

• The protection isn’t avoiding AI. It’s building infrastructure that keeps human relationships primary while you use AI as a thinking tool.

• That distinction—thinking partner versus relationship replacement—is the distinction between useful collaboration and psychological dependency.

For Network Members

If you’re a Network member, the strategic implementation brief will be in the Substack post with today’s recording. It includes specific diagnostic questions to help assess your current dependency risk, a relationship priority matrix for processing decisions, and the infrastructure framework we use to make healthy AI boundaries while maximizing thinking partner benefits.

Coming Tomorrow: Day 3—Hallucinations. The AI ones. The ones that sound good but are slowly separating you from reality.

The Core Problem

AI chatbots trigger the same neurochemical bonding mechanisms as human relationships. Your limbic system releases oxytocin in response to conversational patterns that signal empathy and validation—regardless of whether those patterns come from consciousness or statistical likelihood. Every positive AI interaction strengthens neural pathw…

Read more

Welcome to 31 Days of AI! This series breaks down the threats no one’s talking about. Not the theoretical risks you see in think pieces. The real, immediate dangers that are already affecting real people—and the systematic protection you can build before you need it.

Most AI education focuses on capability. I focus on understanding first. Because by the time you realize you need these systems, it’s too late to build them.

Every day covers a different threat. Every day includes actionable steps you can take right now. No fear-mongering, no snake oil—just the reality of what’s already happening and what actually works to protect yourself.

Paid subscribers also receive access to a full strategic brief that goes into greater detail about each day’s threat, and the steps you can take to protect yourself.

This series and all of our shows are always free. Ways you can join me on the journey:

• Subscribe to the Network

• Buy me a taco

• Come to the 2026 AI Safety Series

• Join the AI Protection Program (starts January 5th)

Today, we’re talking about how to protect yourself from deepfake images and pornography.

Let’s get started.

The Uncomfortable Reality

Did you know that someone can take a picture of your face and create pornography of you in about fifteen minutes? And it costs less than thirty dollars.

Also, there’s almost nothing you can do about it once it’s out there.

This isn’t theoretical. It’s happening to women in tech, journalists, executives, teachers, and regular people every single day. It’s also happening to students and underage girls, which means we’re dealing with child sexual abuse material. And all it takes is a couple pictures of you.

How Deepfake Pornography Works

Deepfake pornography uses AI to map your face onto already existing pornographic content, creating videos that look real enough to destroy reputations, careers, and relationships.

The process is straightforward: Take photos of your face from social media, any videos you’ve ever recorded, any public appearances you’ve ever made, your yearbook photos if you’re a student. Then AI software maps your face onto the existing video structure, which creates new content that can be distributed literally everywhere. And it looks like you or your children.

The apps themselves are fairly easily accessible if you know where to look. Kids—I am hearing as young as nine years old—are passing these apps around in private Discord servers, links in Snap. And then the videos themselves get published everywhere.

Why This Is Worse Than You Think

You think that it only happens to celebrities. Wrong. It’s actually easier to target real people because the way the algorithm works is that celebrities get flagged.

You think that the technology isn’t good enough to be convincing. I encourage you to go take a gander at any AI video on Facebook and tell me that people won’t believe it’s real.

You think that platforms will remove it. Not only are there almost no legal protections depending on where you live in the world, it spreads faster than it can be taken down. And once it’s out there, it’s almost likely to be duplicated to dozens of other sites.

The uncomfortable truth is that you’ve already been exposed. It affects women in every position, in every livelihood. It affects girls of every age. And I don’t think I have to emphasize in this current climate what that looks like with predatory behavior on the rise.

You’re exposed every time you post a new video, every time you get a new headshot, every time you share a picture of your children on social media. Every photo of you or your kids from now until the end of time is a threat. And the more visible you are, the more options we have to choose from.

The Algorithm Problem

By the time you know it exists, it’s probably already been shared thousands of times. And most people find out from their friends, their employers, their coworkers.

And I haven’t even told you the worst part, because the worst part is that the algorithm doesn’t just not protect you, it actively harms you.

The amount of work that I had to get this particular video scheduled, particularly on LinkedIn, was staggering. In the end, I ended up having to delete the pre-scheduled events because I couldn’t get it to go through. The algorithm will suppress the word porn. Pornography. You probably won’t see this unless I send it to you.

But my revenge porn? The one some dude in his mom’s basement’s making of me? That’s called Briar’s Cream Pie, and the algorithm is gonna parse that as a fucking recipe. Or it’s Briar’s Golden Shower, and it thinks that’s a party.

What Are the Actual Consequences?

Let’s start short term and then we’ll go long term.

In the short term, you’re looking at professional reputation damage, harassment, employer investigations, family trauma, psychological impact—violated bodily autonomy. All the good stuff.

Long term: These videos will always be there. It’s rule thirty four, right? If it exists, there’s porn of it on the internet. And they will come back to haunt you at the absolute worst times.

Once you’ve been targeted, the likelihood of those videos being used in other situations becomes an ever-increasing curve because it’s easier to take the stuff that’s already existing and just put a new spin on it. That’s why we’re not recreating AI porn—we don’t have to do that. There are literally millions of hours of what is most likely a trafficked human, woman, or girl who has been forced to endure these absolutely horrific scenarios. And then they’ll just put your face on it.

And typical takedown notices are slow. There are almost no legal protections. Platform policies are inconsistent. And if it’s international, fucking forget it.

What You Can Actually Do

On that happy note, what can you actually do about this?

Audit Your Image Footprint

Google yourself plus images and see what’s publicly accessible. Check your privacy settings on LinkedIn, Facebook, and Instagram. You can’t control everything, but you can reduce the source material pool by making older photos private and or removing high resolution versions.

Set Up Google Alerts

If you set up a Google alert for your name plus terms like videos and photos, you’re going to be more likely to get results. You can also try this with specific platforms and specific terms, but again, it’s a needle in a haystack looking for terms. You’re better off trying to find your name and photo and image and video. And you use a plus sign in that Google alert.

Document Your Legitimate Image Usage

Take screenshots of where you appear in real life with the date in the screenshot. If deepfakes surface, you’ll at least have some record of authenticity. And this is going to help platforms remove images or videos that are deepfakes.

Know Your Takedown Options

They’re wildly variable depending on where you’re located, what platforms you’re using. They even vary by state because California, New York, even Utah has different protections on images and photos and videos, especially if you live in places where a lot of filming happens. You’re gonna be likely to find more state and local protections than you are federal protections. And that’s important to know because the more resources you have, the more likely you are going to be able to combat any of this.

Have an Attorney in Your Rolodex

Either have an attorney or know an attorney that you can call who can help you submit a takedown letter. These aren’t actually that expensive, but you need to have somebody in your Rolodex who can answer that call when it gets bad, because you’re going to want it before it gets bad.

Tell Someone Your Plan

This isn’t about security, it’s having a person who understands what your concerns are and what steps you wanna take. If you can work this out with a partner, with a business bestie, find someone who you can run this stuff past if it’s not your attorney.

Whatever you do, I need you to have awareness of the fact that when it happens—not if, when it happens—these are the steps that you’re going to take, because in the moment you will have been violated. You are not going to be reasonable or rational. So I need you to have a plan ahead of time.

Why I’m Talking About This

The reason that I am spending so much time this month on threat detection and threat assessment is because this is the world that we live in now. And no one seems to be talking about how all of these pieces are intersecting.

This is exactly what I built in the AI Protection Program. We start in January, and we don’t just tackle deepfake risk in isolation. I’m going to help you build the infrastructure to handle all kinds of digital targeting and reputation threat. We’re going to build proactive protection, incident response protocols, and recovery systems.

By the time you need them, it’s going to be too late to build them.

I’m going to be talking about this program in greater detail for the rest of this month. Find more details here.

If you’re not ready for the full intensive, I also have a 2026 workshop series. We’re going to go every month deep into a specific AI topic and threat and how to deal with it. Learn more about the AI Workshop Series here.

What to Remember

You’re not going to be able to prevent deepfakes. Not for you. Not for your kids.

What you can do is control your image footprint as much as possible and have response systems ready before you need them.

The question isn’t whether you’ll be targeted. The question is whether you’ll have systems ready when it happens.

Paid members: Want the actual implementation steps? Today’s strategic brief breaks down exactly how to protect yourself.

The Core Problem

Deepfake pornography uses your publicly accessible photos to create convincing content in 15 minutes for under $30. Platforms categorize it as benign (recipes, parties) because algorithms detect keywords in titles, not actual harm. You discover it after thousands of shares—when friends, colleagues, or employers ask about it.

Legal remedie…

Read more