Artificial intelligence researcher Connor Leahy is sounding a new alarm about the direction of advanced AI, warning that the technology is evolving faster than many people understand — and possibly faster than its creators can control.
In a recent interview titled “AI Is Mutating,” Leahy argued that the public often misunderstands how modern artificial intelligence is built. Many people imagine engineers carefully writing every rule an AI system follows. But according to Leahy, today’s most powerful AI models are trained more like complex systems that are grown, tested, and adjusted, not fully hand-coded from the ground up.
That difference matters. If an AI system is “grown” through massive training runs rather than directly programmed line by line, even the companies building it may not fully understand why it behaves the way it does. This is one of the biggest concerns in the AI safety debate: powerful systems can produce useful, surprising, or even dangerous behavior that was not intentionally designed.
Leahy, who co-founded EleutherAI and now leads the AI safety company Conjecture, has become one of the most outspoken critics of the race toward artificial general intelligence, or AGI. He argues that the current competition between major AI labs is pushing companies to build more capable models before society has solved the alignment problem — the challenge of making sure AI systems reliably follow human values and goals.
The phrase “AI is mutating” does not mean artificial intelligence is alive in a biological sense. It means that AI systems are changing in capability, behavior, and complexity in ways that can feel unpredictable. New models are showing stronger reasoning, coding, planning, image generation, voice interaction, and tool-use abilities. Each jump in capability raises the same unsettling question: do developers truly understand what is happening inside these systems?
Leahy’s warning comes as AI companies continue releasing more advanced models into everyday life. AI is now being used in classrooms, hospitals, hiring systems, financial workflows, military planning, software development, media production, and personal relationships. Supporters argue the technology could accelerate science, cure diseases, improve productivity, and unlock a new era of human progress. Critics argue the same speed could create deepfakes, mass manipulation, job disruption, cyber threats, and systems that become harder to control over time.
One of Leahy’s biggest concerns is that AI companies are financially rewarded for moving fast. The more powerful the model, the more attention, investment, and market value a company can gain. That creates a dangerous incentive structure where safety may fall behind competition. In his view, asking tech companies to police themselves is not enough, because the same companies building the systems also profit from releasing them.
This is why Leahy and other AI safety advocates have called for government action, including stricter oversight of frontier AI labs, liability for harmful systems, and potential limits on the training of the most powerful models. He has previously supported a pause or moratorium on certain advanced AI development until society has stronger safeguards in place.
The debate is not only technical. It is political, economic, and cultural. If AI systems become more capable than humans at persuasion, strategy, hacking, scientific research, or business decision-making, then control becomes the central issue. The question is no longer just whether AI can answer questions or generate images. The question becomes who directs it, who benefits from it, and what happens if its goals do not match ours.
At the same time, not everyone agrees with Leahy’s level of concern. Some AI leaders argue that current systems are still tools, not independent agents, and that safety research is improving alongside capability. Others believe slowing AI development could prevent medical breakthroughs, economic growth, and national security advantages. But even many optimists admit that today’s AI systems remain difficult to fully interpret.
That uncertainty is what makes Leahy’s warning resonate. AI is becoming more powerful, more widespread, and more deeply embedded into society before the public has fully caught up to what it is. Whether people see that as progress or danger, the pace of change is undeniable.
Connor Leahy’s message is simple but unsettling: humanity is building machines it does not completely understand, racing to make them stronger, and hoping control can be figured out along the way. If he is right, the biggest AI story is not just what the technology can do today — it is what it may become next.
