It's rare for the creators of a technology to ask for it to be regulated themselves. When the heads of the world's four largest AI labs sign a warning letter together, that's not marketing: it's a signal to be taken seriously. This article explains their concern without catastrophism, but without downplaying it either.
On 4 June 2026, a letter addressed to the US Congress brought together signatures rarely seen side by side. Sam Altman (CEO of OpenAI, creator of ChatGPT), Dario Amodei (CEO of Anthropic, creator of Claude), Demis Hassabis (head of Google DeepMind, Nobel Prize in Chemistry for his work on AlphaFold) and Mustafa Suleyman (head of AI at Microsoft). Four fierce competitors, united for the duration of a warning.
Their message boils down to one chilling sentence: AI now outperforms PhD-level virologists on many technical laboratory questions. And that is eroding the knowledge barriers that historically made biological weapons inaccessible to the average person.
What exactly are we talking about?
For a long time, building a dangerous biological weapon required extremely specialised knowledge, accumulated over years of training. This rare expertise acted as a natural barrier: even with bad intentions, most people simply didn't have the know-how to follow through.
The signatories' concern is that AI is lowering this barrier. If a model can answer, step by step, the kind of pointed technical questions that only a seasoned virologist used to master, then dangerous know-how potentially becomes accessible to far more people. It's not that AI turns malicious: it's that dangerous knowledge becomes easy to obtain.
Knowing is not doing. Knowing the theory of a pathogen doesn't mean you can produce it: you need labs, materials, practical skills. The letter doesn't say AI will create pandemics tomorrow. It says one of the protective layers, the scarcity of knowledge, is weakening, and that the other layers therefore need strengthening. This is a defence-in-depth logic, not a prophecy.
Their concrete demand: screening DNA
What makes this letter credible is that it doesn't just wave a fear around: it proposes a precise, actionable measure. The signatories are calling for mandatory screening and registration of synthetic DNA and RNA orders.
Explanation. Today, companies manufacture strands of DNA or RNA (the molecules that encode life) to order and ship them to research labs. It's a legitimate service, essential to science. The letter's idea: systematically check what is ordered and by whom, in order to block any order matching a known dangerous agent. It's a practical checkpoint, located not on knowledge (impossible to lock down) but on the physical material needed to follow through.
The letter was also co-signed by key industry players, including executives from DNA synthesis companies such as Twist Bioscience, and leading scientists like David Baker (Nobel Prize in Chemistry) and Martin Hellman (a cryptography pioneer). When the manufacturers themselves ask to be regulated, the risk is judged to be real.
Why this is an important moment
This letter marks a shift in tone for the industry. For years, AI labs have mostly touted the benefits of their technology. Seeing them unite to call for oversight of one of its possible uses shows a form of maturity, or at least lucidity.
One could see cynicism in it, of course: asking for regulation you can influence is also a way of keeping control of the narrative. But on this specific issue, the alignment between competitors, scientists and manufacturers suggests genuine concern. The same companies tear each other apart on everything else, as we see in the distillation dispute between Anthropic and Alibaba. That they agree here says a lot.
What to take away from it
This episode illustrates a truth about AI in 2026: the same capability that heals can also harm. The AI that helps design drugs and understand proteins (like AlphaFold, precisely) is the same one that, misused, could lower the barriers to the worst. Technology is neutral; it's the uses and the guardrails that make the difference.
The good news is that solutions exist and are concrete. Screening DNA orders, training models to refuse dangerous questions, sharing intelligence between labs and states. None is perfect, but stacked together, they form a solid defence. The real danger wouldn't be the technology itself, but inaction in the face of a risk its own creators took the trouble to name. For once we've been warned in advance, we might as well make the most of it.