I get it. You’re stuck on a problem, you fire up ChatGPT for a quick brainstorm. I do it all the time for email drafts, coding snippets, or even just figuring out what to make for dinner. It’s a fantastic thought partner for low-stakes stuff.
But for, you know, running a country? That’s a whole different level of “trust the process,” and honestly, it’s a terrifying one.
Yet, here we are. The Prime Minister of Sweden, Ulf Kristersson, just casually admitted in an interview that he regularly uses AI chatbots like ChatGPT for a “second opinion” on major decisions. My jaw just about hit the floor. He said he uses it to ask things like:
“What have others done? And should we think the complete opposite?”
This is wild. While I’m all for leaders being curious about new tech, this feels less like curiosity and more like outsourcing national strategy to a glorified text predictor owned by a foreign tech giant. And I’m not the only one who thinks this is an absolutely bonkers idea.
🚨 The Alarm Bells Are Ringing
Almost immediately, experts in AI and political science started waving giant red flags. Virginia Dignum, a professor of responsible AI, put it perfectly:
“The more he relies on AI for simple things, the bigger the risk of overconfidence in the system. It is a slippery slope.”
Her closing line was an absolute mic drop and should be plastered on the wall of every government office on the planet:
“We didn’t vote for ChatGPT.”
It’s a simple but profound point. Democratic accountability is the bedrock of a free society. We elect leaders to use their judgment, guided by their teams of human experts who can be questioned, who can show their work, and who are accountable for their advice. When you start feeding state-level thoughts into a black box algorithm, that entire chain of accountability shatters.
⚙️ Let’s Break Down Why This is a TERRIBLE Idea
This isn’t just a weird quirk; it’s a multi-faceted risk to national security, policy integrity, and democratic principles. Let’s unpack the biggest dangers here, because they apply to any organization, not just a whole country.
- 📌 The Security Nightmare: An AI consultant, Jakob Ohlsson, called the Prime Minister’s approach “amateurish,” and he’s not wrong. A spokesperson for the PM insisted that “sensitive information” is never used. But that’s a dangerously naive view of how intelligence works. You don’t need to upload a classified document to leak state secrets. A series of seemingly innocuous prompts about economic policy, infrastructure priorities, international relations, and social issues can be pieced together by a competent analyst. Over time, these prompts create a detailed mosaic of a government’s strategic thinking, its blind spots, and its potential future actions. And where does all that data go? Straight to the servers of a US-based tech company. A company that, as Ohlsson pointed out, is becoming more powerful than many states, in a country with a deeply uncertain political future. You’re essentially handing the keys to your strategic kingdom over to a foreign commercial entity.
- 💡 The Hallucination Hazard: We’ve all seen AI get things hilariously wrong. It confidently makes up facts, invents historical events, and creates fake legal precedents. We call these “hallucinations.” It’s funny when ChatGPT invents a recipe that calls for a cup of gravel. It’s a national security crisis when a Prime Minister asks for historical context on a border dispute and the AI invents a treaty that never existed, potentially influencing a real-world diplomatic incident. These models are designed to be plausible, not truthful. They’d rather give you a confident, well-written lie than admit they don’t know something. Relying on them for a “second opinion” on factual matters is like asking a compulsive liar to be your fact-checker.
- 🧠 The Deskilling of Expertise: Writer Signe Krantz hit the nail on the head when she asked why the PM would use “random number generators” for advice instead of “his large and well-paid staff of experts.” This is the danger of over-reliance. When you have a tool that gives you an easy answer, you stop developing the critical thinking muscles needed to find the hard answer. Human experts bring decades of experience, nuance, cultural understanding, and intuition to the table. They can read between the lines. An AI can’t. It flattens every problem into a dataset. Worse, as Krantz noted, it can often take an expert longer to identify and fix the subtle but critical mistakes an AI makes than it would have taken them to do the job right from the start. It’s not a shortcut; it’s a detour through a minefield.
- ❓ The Accountability Black Box: This might be the most crucial point for any democracy. If a human advisor gives bad advice, they can be held accountable. They can be questioned. They have to explain their reasoning, cite their sources, and defend their position. If an AI gives advice that leads to a disaster, who is to blame? The Prime Minister who acted on it? The engineers who wrote the code? The dataset it was trained on? Sam Altman? The system is fundamentally a “black box.” We don’t truly know why it produces a specific output. It’s a complex web of probabilities and weights, not a transparent line of reasoning. You can’t have governance without accountability, and you can’t have accountability with a black box.
✨ But Isn’t This Just Forward-Thinking?
Of course, there are some who defended the Prime Minister. One editorial suggested he “is absolutely right about AI technology,” but with a massive caveat: that he obviously knows not to actually trust the answers. This is a pretzel of logic. If you know the tool is unreliable, why are you consulting it for some of the most important decisions a person can make? It’s like saying, “I consult my Magic 8-Ball for investment advice, but of course, I know its answers are random.” It just doesn’t make sense.
This whole episode is a perfect symptom of the current AI hype bubble. We’re being sold a vision of an inevitable, awesome AI-powered future by people who stand to make billions from it. But the reality on the ground is a mess of security holes, torrents of AI-generated slop, and systems that are nowhere near as intelligent as the marketing suggests.
✍️ My Take: A Tool, Not a Throne
Look, AI is a game-changing tool. It’s supercharging creativity, automating tedious work, and accelerating scientific discovery. But a tool is only as good as the person using it, and it must be the right tool for the job. You wouldn’t use a hammer to perform surgery.
Using a public-facing LLM for high-stakes government decision-making is not just the wrong tool; it’s an act of profound recklessness. It’s a security risk, a reliability nightmare, and an affront to the principles of democratic accountability.
So, how should we actually use these powerful tools responsibly?
- ✅ Brainstorming Only: Use AI as a creative spark. Ask it for ten ideas on a topic, but then use your own expertise to evaluate, discard, and refine them.
- ✅ NEVER Input Sensitive Data: This is non-negotiable. Assume anything you type into a public AI tool could end up on a billboard in Times Square. Your personal info, company secrets, and especially national strategies do not belong there.
- ✅ Fact-Check Everything: Treat every output as a first draft written by a brilliant but untrustworthy intern. Verify every fact, check every source (which the AI often invents), and challenge every conclusion.
- ✅ Understand Its Limits: Remember, it’s not thinking. It’s a pattern-matching machine that predicts the next most likely word. It has no understanding, no consciousness, and no ethics.
Artificial General Intelligence might be a real thing someday. But it isn’t today. What we have now is an incredibly powerful, but flawed, technology. It’s Eliza with internet access, and that’s not something any world leader should be taking advice from.
The European Union’s AI Act, which Prime Minister Kristersson has criticized, is a landmark legislative effort to create a common regulatory framework for artificial intelligence. It proposes classifying AI applications by risk level, from unacceptable to minimal, and imposing stricter rules on high-risk systems used in areas like critical infrastructure, law enforcement, and democratic processes. The goal is to ensure AI is safe and respects fundamental rights while fostering innovation.
Concerns about AI bias, as highlighted by academics, stem from the data used to train the models. AI systems learn from vast amounts of text and images, often scraped from the internet. If this data reflects historical societal biases, stereotypes, or underrepresents certain groups, the AI can perpetuate and amplify these flaws, leading to unfair or discriminatory outputs.
While the Swedish case is notable, governments worldwide are exploring AI. The UK has used AI to summarize public consultations, Singapore employs it for transport management, and the U.S. has issued executive orders on the trustworthy use of AI in the public sector. These examples show a global trend toward leveraging AI for efficiency, but also highlight the universal challenge of establishing ethical guardrails and maintaining public trust.