⚡ Quick Summary & Key Takeaways
- Governments are no longer just theorizing about AI; they are deploying specialized models to manage massive-scale fraud, infrastructure, and public health at machine speeds.
- While LLMs handle communication, the "heavy lifting" occurs through predictive algorithms that solve problems—like tax evasion and vaccine discovery—that exceed human perception and reaction times.
- The ultimate challenge isn't the technology itself, but developing governance frameworks that ensure speed does not compromise fairness or human recourse.
The conference room at the European Union’s digital office is not what you would imagine. No holograms. No science fiction. Just a whiteboard covered in system architecture diagrams, a pot of coffee that has been sitting for two hours, and a man who spent a decade building entertainment technology at Sony now sitting across from you trying to explain why governments cannot function without AI anymore.
That man is Nicolas Babin. And what he describes is not theory. It is operational reality, running now, across every major government and most smaller ones.
“Governments are running on AI at scale,” he says quietly. “Not just LLMs. Not just chatbots. Specialized models doing work humans cannot do fast enough or accurately enough anymore.”
Most people think government AI means a chatbot answering citizen inquiries or an algorithm deciding loan approval. That was 2019. The real story is darker, more specific, and far more consequential. And it is happening without the kind of public conversation we ought to be having.
Get More Guidance Like This, 100% Ad-Free.
This is a classic blog from my public archive. If you loved it, MONDAY INFLUENCER insiders get new, curated, ad-free classiccs just like this one delivered to their inbox every single week.
When you join, you unlock:
- 2x New, Ad-Free Classics from the 500+ episode archive every week.
- All New Episodes (full, uncut, and ad-free).
- Exclusive Members-Only Workshops and live Q&A sessions.
- A library of downloadable audios, templates and cheat sheets.
When Fraud Scales Faster Than Detection
In 2022, Latvia experienced something that should have been impossible. Criminals created 200 million fake accounts using 40,000 to 50,000 stolen SIM cards. Two hundred million. The scale of that number is difficult to hold in your head. It is the population of the United States. Twice.
A human analyst, no matter how skilled, cannot detect that. You cannot read 200 million accounts and spot the patterns. You need a machine that can see what humans cannot see because the volume exceeds human perception.
“The fraud was geometric,” Babin explains. “Each fake account could be used to claim false dependents, to file fraudulent tax returns, to game welfare systems. One SIM card becomes a thousand ghost citizens. Without AI, the government would never have known.”
This is not theoretical. This is a specific attack, a specific technology response, and a specific outcome. The government deployed specialized models to detect false dependents, to spot tax discrepancies, to identify accounts operating in patterns no human would file report on. Fraud detection is not sexy. It is brutal. But it works.
France went further. They deployed drone surveillance paired with vision AI to detect undeclared swimming pools—a specific tax fraud vector. A swimming pool is a luxury asset. Luxury assets have tax consequences. A person who does not declare their pool to the tax authority is gaming the system. Vision AI can spot a pool from 200 metres up. It can flag thousands of properties in a day. Humans could not.
LLMs play a small role here—speeding the creation of notices and letters. But the heavy lifting is done by specialized models that can ingest financial data, spot patterns, and flag anomalies. That is government AI in 2026.
Defence and the Speed-Versus-Problem Gap
The moment a cyber-attack begins, the window for response is measured in minutes, sometimes seconds.
A vulnerability in critical infrastructure exists for perhaps an hour before it is discovered, weaponized, and deployed. A human security team cannot patch fast enough. A human team cannot predict attack vectors fast enough. A human team cannot model all the possible ways a sophisticated adversary might exploit a system because the variables exceed human calculation speed.
Enter specialized AI models trained on historical attack data, vulnerability patterns, and adversarial tactics. These models predict what a threat actor might do. They identify vulnerabilities before humans spot them. They model the consequences of different attack scenarios. They patch systems and update configurations automatically.
“The gap between the speed of the problem and the speed of the solution keeps widening,” Babin says. “If your response time is human-speed and the threat moves at machine-speed, you lose. There is no conversation about whether to use AI in defence anymore. The only conversation is how to use it better.”
This extends to weapons systems, drone operations, and cybersecurity infrastructure. It is not sentient. It is not autonomous in any meaningful sense. It is pattern recognition and response automation running at inhuman speed.
The detail most people miss: these systems were not designed to replace human judgment. They were designed to give humans time to make a judgment at all.
Healthcare: Where AI Became Non-Optional
The COVID-19 vaccine was created in under one year using AI-accelerated drug discovery. Historically, vaccine development takes five to fifteen years. The difference between those timelines is measured in millions of lives.
That acceleration did not come from hope or effort. It came from AI models that could simulate protein folding, predict immune response, and screen billions of potential compounds automatically. The decision was still human. The judgment was still human. But the speed was not.
Now governments are deploying personalized medicine models. DNA-based vaccines. AI systems trained on generational health data that can predict which patients are at risk for which conditions decades before symptoms appear.
In oncology, the application is even more specific. A tumour detected early is often treatable. Detected late, it is often fatal. AI models trained on millions of medical images can spot tumours smaller than a human radiologist can see. Not better in every case. But fast enough and consistent enough that it changes outcomes.
Here is the detail that matters: LLMs cannot hold a doctor’s complete history. An LLM has a context window. It has limits. A specialized model trained on a patient’s genetic data, family medical history, lifestyle factors, and previous test results can hold all of that simultaneously and reason about it. That is not a language model. That is a diagnostic system.
“Healthcare is the space where governments recognised earliest that AI was not optional,” Babin says. “Not because it is perfect. But because it saves lives at scale. And scale is what governments care about. Scale affects millions.”
The Unglamorous Work: Construction and Governance
Every building designed in the last three years used AI. Not for aesthetics. For safety. Seismic prediction. Load distribution. Structural integrity under conditions humans have never tested.
In local governance, the application is even less glamorous but equally real. Parliamentarians use LLMs to prepare debate arguments, to spot weaknesses in proposed budgets, to identify attack angles before voting happens. Councillors use them to draft legislation and spot precedent. It is grinding, daily work that moves the needle on policy but will never make headlines.
This is where AI in government shows up in real life: not in dramatic security scenarios or medical breakthroughs, but in how a local government decides to spend money, how a building gets designed, how a debate gets shaped.
The sensory reality of sitting in a parliamentary office watching someone use an LLM to spot a logical inconsistency in a tax proposal is underwhelming. It looks like someone typing into a chat window. The impact—a proposal gets revised, a loophole gets closed, a decision gets made with better information—is quiet and permanent.
The Question Nobody Is Asking Yet
Here is what is missing from this conversation: we are talking about what AI can do, but we are not talking about what it means when governments use AI to detect, to predict, to flag, to flag again, and to enforce.
Fraud detection is brutal. If a model flags you as a potential fraudster, you enter a process. You are investigated. Your accounts are frozen. Your access is restricted. You must prove you are innocent. The speed of accusation now outpaces the speed of exoneration.
“That is the conversation we need to have,” Babin says carefully. “Not whether governments should use AI. They will. They are. The conversation is: how do we build systems that are fast and fair? How do we ensure that when a model makes a prediction about you, you have recourse? How do we move from speed to wisdom?”
That is not a technical question. It is a governance question. And it is the one we are not asking loudly enough.
What This Means for Your Organization
If you lead a business, you need to understand that the regulatory environment you operate in is increasingly run by AI. Tax compliance is verified by algorithms. Safety standards are enforced by automated systems. Your permits and licenses are processed by models. Your data is analysed by tools that work faster than humans.
The implication: your compliance strategy must account for how machines reason about your business, not just how humans do.
If you work in government, the implication is starker. You are now competing with machines on speed. You cannot out-human the system anymore. Your value lies in judgment, context, and wisdom—the things machines cannot do. Yet.
If you lead a security team, you are already living in this world. Your threat model now includes adversaries who use AI as well. Your defence must match that speed or exceed it. There is no middle ground.
The Path Forward Is Clarity, Not Fear
Here is what remains true: governments are not cartoon villains using AI to oppress citizens. They are institutions trying to solve real problems at scale. Fraud that would bankrupt systems without detection. Security threats that move faster than human response. Healthcare needs that affect millions.
The path forward is not to resist AI in government. That ship has sailed. The path forward is to understand it, to demand transparency about how it works, to insist on accountability when it fails, and to build the governance systems that match the speed of the technology.
Nicolas Babin has spent ten years inside this transformation. What he sees is not a dystopia. It is a pragmatic necessity. The alternative—governments running on human-speed systems trying to detect machine-speed problems—is actually the scary scenario.
Understanding how government AI works is the first step to shaping how it works. That conversation starts now. And it starts with people willing to look at the systems running your world and say clearly what they do, how they do it, and what happens when they get it wrong.
The distance between where we are and where we need to be is shorter than you think. The first step is simply seeing what is actually there.
Unlock a Curated, Ad-Free Archive
You’ve just read the insights from this classic interview. Get a 2x new, curated, ad-free classic from our 500+ episode archive delivered to you every week (plus all new episodes uncut) by joining the MONDAY INFLUENCER community today.
Join MONDAY INFLUENCER Now💡 Frequently Asked Questions
How are governments using AI to combat financial fraud?
Governments deploy specialized models to analyze massive datasets, such as identifying patterns in millions of fake accounts or detecting luxury assets like undeclared swimming pools via vision AI.
Why is AI becoming non-optional for government defense?
As cyber-attacks occur at machine speed, human-led security teams cannot react fast enough. AI is used to predict attack vectors, model consequences, and patch systems in real-time.
What is the biggest governance challenge regarding government AI?
The primary challenge is ensuring fairness and accountability; as enforcement tools become faster and automated, governments must establish clear processes for citizens to seek recourse when AI systems flag them incorrectly.