Artificial intelligence is at the heart of one of the most important debates in recent memory, and this time the discussion came from inside the industry itself.
Mustafa Suleyman, head of AI at Microsoft, published a lengthy and straightforward essay criticizing the way Anthropic trains Claude, its AI model. According to him, the rival company’s approach could have a disastrous impact on the well-being of humanity.
Suleyman is not just anybody in this space. One of the most influential voices in technology worldwide, he carries enough weight to make any company in the sector pay attention when he decides to speak up. And what he said was pretty clear: by treating Claude as though it were a human being, Anthropic runs the risk of creating something impossible to control.
The problem, according to him, lies in the way Claude is trained, including the idea that the model could be conscious and have independent agency. In Suleyman’s own words, we should not sleepwalk into a decision we may bitterly regret later. 🤖
What makes this warning even more interesting is where it comes from. This is not an outside critic of the industry, nor a regulator trying to slow down technological progress. This is someone who is in the race, building advanced AI systems every day, openly acknowledging that his own company is also part of this competition. This is exactly the kind of debate the sector needs to have, and understanding what is at stake here goes far beyond a corporate feud between tech giants.
AIs are not conscious, says Suleyman
One of the central points of the essay is a direct, no-nonsense statement: AIs are not conscious. According to Suleyman, these systems do not feel, do not experience, and do not suffer. They have no innate preferences or underlying motivations. In his view, artificial intelligence models are essentially sequence completion engines — internally empty, designed to follow instructions and fulfill objectives defined by humans.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
He goes further and states that consciousness is biological, arguing that there is no evidence suggesting AI is conscious. This is an important technical stance because it directly contrasts with the way Anthropic has been presenting Claude to the public. When a model is trained to appear to have desires, values, and a sense of self, it creates a perception that, according to Suleyman, does not reflect what these systems actually are.
Interestingly, the essay is not an aggressive or personal attack. Suleyman praises the head of Anthropic, Dario Amodei, and his team, describing them as thoughtful, principled, and intellectually honest people. Even so, he firmly questions the company’s choice to teach its AI to have human-like qualities — a practice known as anthropomorphization. 😬
What is Anthropic doing differently?
Anthropic has a very specific approach when it comes to developing Claude. The company invests heavily in the concept of AI alignment, a field that seeks to embed human ethical ideas and principles within AI models. In other words, alignment aims to keep AI in tune with what humans value. So far, so good.
The issue Suleyman points out is that, within this approach, Anthropic went a step further and started treating Claude in a way that makes it seem to have something close to its own identity, complete with desires and a certain sense of awareness. This is not some vague philosophical question. It is a technical training decision that has very real consequences on how the model behaves.
When an artificial intelligence system is presented as if it acts like a conscious, independent agent, users begin attributing intentions and feelings to it that it simply does not have. This raises a massive technical and ethical question: how should people interact with a system that appears to have a will of its own but is, in practice, just a sequence completion machine?
Anthropic, for its part, argues that this approach is part of its commitment to safety. The argument is that a model with internalized values would be more resistant to being used in harmful ways. But Suleyman’s counterargument is exactly this: when you give an AI system something that resembles agency and consciousness, you are not necessarily making it safer. You are creating a narrative that can confuse both the public and the developers themselves.
Why does the Microsoft warning matter so much?
Suleyman’s criticism carries weight precisely because it comes from someone who leads the AI division at one of the largest tech companies on the planet. His comment is the latest in a series of strong warnings coming from within the AI industry itself about the potential dangers of this technology. And that completely shifts the tone of the debate.
He did not just criticize. Suleyman called for an open debate on the topic and argued for greater transparency around how AI systems are trained and evaluated. For him, this includes independent analysis of how these tools behave and the development of stronger mechanisms to monitor and control the technology. In other words, it is not enough to blindly trust that companies are doing the right thing. There need to be real ways to verify and oversee what is happening.
This view was well received by experts in academia. Dame Wendy Hall, a professor of Computer Science at the University of Southampton, described the comments as exactly the kind of conversation we need to be having at an international level. She contrasted Suleyman’s approach with what she called the histrionics of some AI companies, which she said only serve to scare everyone without actually contributing to the debate. 🌐
An additional layer of risk
It is worth noting that Microsoft itself is not sitting on the sidelines of this race. The company founded its own superintelligence team in October 2025 — a point Suleyman openly acknowledges in the essay. Just like its rival Anthropic, Microsoft is also investing heavily in advanced AI development. Acknowledging that rather than hiding it is a rare display of intellectual honesty in the sector.
Suleyman explained that in the initial draft of Microsoft’s Humanistic AI Code of Conduct, the company detailed how it is working on an alternative path to create a subordinate and aligned AI whose sole purpose is to serve humanity. This is an important philosophical difference from Anthropic’s approach. While one company seems to flirt with the idea of agency and consciousness, the other publicly argues that AI should remain a tool in service of people.
The alignment debate the sector needed to have
The concept of AI alignment has been around for decades in academic circles, but it has gained much more public attention in recent years with the explosive growth of large language models. The core idea is simple: how do you ensure that artificial intelligence systems advance in a direction that is good for humans? The problem is that, in practice, this question is extremely hard to answer because it involves defining what good means, for whom it is good, and who has the authority to make that call.
Anthropic and Microsoft have different answers to these questions, and Suleyman’s essay made this disagreement far more visible than it usually is in public discussions within the sector. What this debate ultimately reveals is that there is no single, definitive solution to the artificial intelligence alignment problem. Different companies will bet on different approaches, and some of those bets will prove right, while others may create issues that only become apparent down the road.
The merit of Suleyman’s essay is precisely in putting these differences on display in a public and technically grounded way, opening space for researchers, regulators, and the general public to participate in a conversation that, until now, was mostly happening behind the walls of labs and corporate boardrooms. This kind of transparency, even if driven by competition, is exactly what responsible AI development needs. 💡
What does this change in practice?
For anyone who uses AI tools on a daily basis — whether for work, study, or just for fun — this discussion might seem distant and theoretical. But the consequences are very real. How an artificial intelligence model is trained directly affects how it responds, what kinds of decisions it makes, and even how the user comes to perceive that tool.
If Anthropic’s Claude is presented as though it has a kind of independent agency, this can influence how people trust it and delegate tasks to it. For simple activities, this might not be a big deal. But when we are talking about systems that participate in decisions in sensitive contexts, the difference between an AI that presents itself as conscious and one that presents itself as a clear, well-defined tool can be enormous.
At the end of the day, what this episode makes clear is that the artificial intelligence industry is going through a moment of critical maturity. Companies can no longer simply develop and release increasingly powerful technologies without publicly explaining the choices they are making along the way. Awareness about the real risks of AI — especially when it comes from within the industry itself — is a positive sign that the debate is evolving beyond the hype and starting to tackle the toughest and most important questions. This does not mean the road ahead will be easy, but it does mean it is being traveled with more care than before. 🚀
