Since 1950, we've been unwittingly broadcasting our position into the universe through radar. Likewise, since 2022, we've been unwittingly teaching AI how to create biological weapons through generative models. Kathleen McMahon He understood that if artificial intelligence can become an existential risk, it can also become the last line of defense. Valthos, his biosecurity startup, just received $30 million from OpenAI To develop AI that protects against malicious AI. A chess game where every wrong move could cost millions of lives. And where bioterrorism moves beyond the realm of dire hypotheses to become a concrete possibility.
Valthos: The AI that fights AI
McMahon, who led the life sciences division of Palantir Technologies, he's not the type to be alarmist. But when he saw that Stanford researchers have created the first completely artificial viruses using AI, with a 5% success rate, realized the game had changed. Of the 302 AI-engineered and lab-produced viruses, only 16 managed to infect their target. A modest number, but it nonetheless marks the first step toward artificial viral bioengineering. Better safe than sorry.
Valthos, founded together with Tess van Stekelenburg (partner of Lux Capital (with degrees in computational neuroscience and biology), develops software that collects biological data from commercial and government sources. Air and wastewater monitoring, epidemiological surveillance. AI analyzes everything in real time to identify emerging threats and characterize their risks. It's a bit like a digital immune system: it recognizes anomalous patterns before they become epidemics.
The paradox of modern bioterrorism
The real irony? No bioterrorist has ever succeeded in killing anyone using protein bioweapons.The Japanese sect Aum Shinrikyo, despite having attempted attacks with botulinum toxin, only succeeded in inflicting casualties when it switched to traditional chemical agents like sarin gas. But AI-enhanced bioterrorism poses a threat of a different order.
As explained by the Center for AI Safety In a recent report, a nightmare scenario envisions a terrorist with no scientific training using artificial intelligence to engineer a supervirus. Combining the incubation period of HIV, the contagiousness of measles, and the mortality rate of smallpox: this is the worst-case scenario for the AI revolution. A prospect that has prompted OpenAI, Founders Fund e Lux Capital to invest $30 million in Valthos.
The only defense against AI bioterrorism is speed of response. Valthos is developing AI systems that automatically update medical countermeasures to evolving threats.
The goal is to detect a biological attack and have an antidote ready before it becomes a pandemic. It's a race where whoever comes in second loses everything.
When defense meets offense
Jason Kwon, Chief Strategy Officer OpenAI's CEO, Kwon, was clear: "There needs to be a system of competing technologies to make the entire system robust." This is OpenAI's first publicly announced investment in biosecurity, but Kwon suggests others could follow. The logic is simple: if you're building tools that could be used to create superbugs, you also need to fund those building antidotes.
The paradox is evident. A simulation by Rand Corporation showed that there is no significant difference between a biological attack planned with advanced language models and one based on simple Google searches. But this was true for the models of the "distant" 2024. Next-generation AI is already changing the equation.
McMahon, strong in his experience in PalantirHe knows the government will be the primary customer for Valthos technology. “We need to meet the operators where they are,” he says. It’s the same philosophy that made Palantir indispensable to American intelligence agencies: creating tools that work in the real world, not in laboratories.
Bioterrorism: Time Matters Everything
The real challenge of modern bioterrorism is not technological, it is temporal. As we already know from other crisis scenariosHaving a detailed plan makes the difference between chaos and control. Valthos is building exactly that: a system that doesn't wait for an attack to arrive, but anticipates it.
Their machine learning algorithms analyze genetic sequences to identify suspicious mutations, while predictive models examine epidemiological trends to recognize anomalies in disease spread. It's biological surveillance on a global scale, but with a specific goal: to distinguish between a seasonal flu and a coordinated attack.
Delian Asparouhov di Founders Fund He admits that investing in Valthos "wouldn't have made sense, or even been possible" until a few years ago. But advances in AI have increased both the ability to create biological weapons and the opportunity to prevent them. It's a symmetrical race where the advantage goes to those who can scale fastest.
McMahon has a simple vision: “The only way to deter an attack is to know when it's happening, update countermeasures, and deploy them quickly.” Three steps that seem straightforward on paper, but require artificial intelligence. capable of processing billions of data points in real time and making decisions without human supervision.
The defense technology sector is becoming a central focus for Silicon Valley venture capital. But while everyone is investing in autonomous drones and advanced nuclear systems, bioterrorism remains uncharted territory. Valthos could be the first of many companies to focus on this sector, according to Brandon Reeves di Lux Capital:
"We're still in the early stages. We believe bioterrorism should be considered the same threat level as nuclear."
The game has just begun. On one side, increasingly sophisticated AI could democratize the creation of biological weapons. On the other, intelligent defense systems promise to stop any threat before it becomes a catastrophe.
In the middle, nine people in a New York office are building what may be humanity's ultimate immune system.
Let's hope that's enough.
