Part 1 of this series looked at what happens when large systems, entire “countries of geniuses,” fall into the wrong hands: a government, a corporation, an institution. Part 2 looks at a much smaller, and in some ways scarier, version of the same problem: what happens when the wrong hands belong to just one person.
A New Kind of Access
Certain kinds of dangerous knowledge have always existed, but they stayed out of reach for almost everyone. Understanding how to engineer or mutate a virus, for example, has traditionally required years of specialized training. The people who hold that knowledge today are elite scientists with PhDs, highly intelligent, and, in nearly every case, far too invested in their careers and reputations to misuse it.
That barrier has acted as a kind of natural safety net. Dangerous knowledge existed, but it was locked behind years of education that filtered out almost everyone except serious, credentialed researchers.
AI puts pressure on that safety net for the first time. A capable enough AI system can act like a knowledgeable advisor, walking a person through complex technical steps they could never have worked out alone. That does not mean any one tool is currently capable of turning an untrained person into a weapons expert. It means the gap between “trained expert” and “motivated individual with the right AI assistance” is shrinking, and shrinking gaps are worth paying attention to before they close completely.
Why This Is Different From a Typical Lone Actor
A lone attacker with a firearm, however awful the outcome, has a limited reach. There is a physical ceiling on how much harm one person with one weapon can cause before being stopped.
A biological weapon does not have that ceiling. A pathogen, once released, does not stay under the control of the person who created it. It can spread far beyond its original target, across cities, countries, and continents, following its own path rather than the attacker’s intentions. That is what makes this category of risk fundamentally different from most other lone actor threats: the damage is not bounded by what one person can physically carry out.
Why This Is Even More Alarming Than Nuclear Weapons
Nuclear weapons are rightly treated as one of the most serious threats humanity has ever created, but they come with built-in obstacles. Building one requires a large, coordinated team, rare materials like enriched uranium, specialized facilities, and extensive testing. Those requirements make nuclear weapons difficult for any small group, let alone one person, to produce in secret.
Biological weapons do not share those obstacles in the same way. In principle, the kind of experimentation involved can happen in a small laboratory setting, without the massive infrastructure a nuclear program demands. That difference in scale is exactly why many researchers consider AI-assisted bioweapons development a more urgent near-term danger than nuclear proliferation. Fewer resources stand between intent and outcome.
What Can Actually Be Done
None of this is a reason to treat AI development as a lost cause. It is a reason to take safeguards seriously before the gap closes further.
One proposed approach is building AI systems with strict internal rules, sometimes called Constitutional AI, so that a model refuses to help with anything related to creating or improving dangerous pathogens, no matter how a request is phrased or disguised. Alongside that, researchers and policymakers argue for firm restrictions on how sensitive biological information is shared and accessed more broadly, not just through AI, so that dangerous knowledge does not become easier to obtain simply because a new tool exists to explain it.
Where This Leaves Us
This is a hard problem precisely because it is not really about the technology being “evil.” It is about a longstanding safety net, the years of training that once stood between dangerous knowledge and dangerous intent, becoming thinner than it used to be. Part 3 of this series continues to look at this same period of uneasy growth, where capability is arriving faster than the guardrails meant to contain it.
