Many AI architects and tech leaders, when asked privately what risk worries them most, give a strikingly similar answer, according to Axios and its “Behind the Curtain” column: the possibility that AI could be misused to help create a deadly pathogen capable of spreading widely, quickly and without early detection.
The report says the companies pushing frontier model development are also grappling with a deeper fear about where those systems could lead if used with malicious intent. Axios wrote that when it privately asked nearly every AI architect and leader about the danger they feared most, the responses were unusually consistent: a biological threat that could move too fast to stop.
A survey put biological danger among the biggest AI risks
A recent survey from MIT FutureTech and the University of Queensland questioned 272 researchers and ranked 24 major AI risks. The results showed that, without mitigation or protective measures, the probability that dangerous AI capabilities could lead to catastrophic consequences by 2030 was above 20%.
The survey defined a catastrophic outcome as more than 1 million deaths or $100 billion in damage. It also found that even with mitigation in place, the chances of catastrophe and of AI enabling weapons capable of causing mass harm were still about 12% each.
Experts warn AI could speed up the path to engineered pathogens
Axios said experts have reached a rare consensus on this point. Advanced AI models, in their view, could make it easier for malicious actors to create deadly pathogens before governments and industry build effective defenses. As AI systems gain a deeper understanding of human biological vulnerabilities, the industry increasingly sees that threat as a realistic possibility.
One concern is that well-resourced bad actors could use powerful models to map human biological weak points and identify genetic modifications that might make known pathogens spread more easily or evade existing treatments. Trained on large volumes of biological and genomic data, AI could generate workable candidates at a pace comparable to hundreds of top scientists working at extreme speed without rest, the report said.
Those designs could then be synthesized with gene-editing tools that are becoming cheaper over time. If a novel pathogen started spreading before public health systems detected it, containing it could prove harder than during the COVID-19 period, according to the report.
Open letter from tech leaders pushed safeguards into focus
Axios also pointed to a public letter signed last month by OpenAI’s Sam Altman, Anthropic’s Dario Amodei, Google DeepMind’s Demis Hassabis, Microsoft’s Mustafa Suleyman and Meta’s Alexandr Wang. The letter warned about biological weapon risks tied to AI and called for stronger safeguards.
That has also sharpened debate over whether open-source models are more vulnerable to misuse outside regulatory controls. The question now at the center of the discussion is how to balance technological progress with protection against malicious use.