The risk of AI enabling biological weapons is not hypothetical
Two men who have spent careers inside the machinery of national security — one who advised presidents, one who guarded critical infrastructure — sat before a national audience and said what many had hoped remained theoretical: artificial intelligence has already narrowed the distance between knowledge and catastrophe in the realm of biological weapons. Their warning was not a forecast but a present-tense assessment, arriving at a moment when governments are still debating whether and how to govern technology that does not wait for consensus.
- A senior AI adviser with White House and Johns Hopkins credentials declared on national television that AI-enabled bioweapons development is no longer a hypothetical — it is a live security threat.
- The same systems racing toward medical breakthroughs can be redirected toward mass harm, and the gap between those two outcomes is shrinking faster than regulation can move.
- Former CISA director Chris Krebs amplified the alarm, drawing on years of protecting critical infrastructure to frame AI as the ultimate dual-use technology — one that strengthens and endangers simultaneously.
- Policymakers now face a knot with no clean thread: regulate too slowly and the risk compounds; regulate too aggressively and innovation stalls, potentially ceding ground to actors with fewer scruples.
- The debate has shifted — no longer whether AI poses a genuine national security risk, but whether governments can design oversight frameworks fast enough to matter.
Ben Buchanan came to 'Face the Nation' carrying the kind of authority that accumulates across a career advising presidents, teaching at Johns Hopkins, and counseling Anthropic — one of the world's most consequential AI companies. He was there to say something uncomfortable: artificial intelligence could help someone build a biological weapon, and that possibility is no longer theoretical. It is happening in the present tense.
Beside him sat Chris Krebs, who ran CISA under the first Trump administration and spent years mapping the ways powerful systems fail and get exploited. Together, they represented a rare convergence — two people with deep government experience, speaking from different institutional vantage points, arriving at the same conclusion.
The core tension they named is one that resists easy resolution. AI is advancing at a pace that outstrips the policy process. The same language models and research tools designed for beneficial purposes carry within them the capacity for misuse — not as a distant risk, but as an accessible one. Buchanan's framing was deliberate: when someone who has spent a career inside national security says a threat is 'not hypothetical,' the word choice is doing real work.
Krebs brought the infrastructure lens — the understanding that dual-use technology doesn't announce which direction it will cut. AI accelerates medical research and weapons development with equal indifference to intent. It hardens cybersecurity and enables attacks. The capability is neutral; the consequences are not.
What gave the conversation its weight was not novelty — security researchers have raised these concerns before — but the clarity and platform with which two credentialed insiders chose to say: this is real, this is now, and the regulatory question is no longer whether but how. The challenge facing policymakers is designing governance that can keep pace with a technology that does not pause to be governed.
Ben Buchanan sat across from Margaret Brennan on "Face the Nation" with the kind of certainty that comes from having advised presidents. He had worked in the Biden administration. He teaches at Johns Hopkins. He advises Anthropic, one of the world's largest artificial intelligence companies. And he was there to say something that sounded like science fiction but wasn't: artificial intelligence could help someone build a biological weapon. Not might. Could. And it wasn't hypothetical anymore.
Buchanan wasn't alone in this assessment. Beside him was Chris Krebs, who had run CISA—the Cybersecurity and Infrastructure Security Agency—during the first Trump administration. Krebs had spent years thinking about what breaks, who breaks it, and how to stop them. Both men had worked at the highest levels of government. Both understood the machinery of national security. And both were now saying the same thing: this is a problem we need to talk about seriously.
The conversation centered on a question that has no easy answer: Should the government regulate artificial intelligence? And if it does, how? The tension is real. AI is advancing rapidly. Companies are racing to build bigger, faster, smarter systems. But those same systems can be used for purposes their creators never intended. A language model trained to write poetry can also write instructions. A system designed to help with research can help with the wrong kind of research.
Buchanan's point was specific: the risk of AI enabling biological weapons development is not a distant concern or a thought experiment for ethics seminars. It is a concrete security threat that exists now, in the present tense. The technology is advancing. The knowledge is increasingly accessible. The gap between capability and catastrophe is narrowing. When someone with his background says something is "not hypothetical," the weight of that statement carries the authority of someone who has spent his career thinking about what actually happens when bad actors gain access to powerful tools.
Krebs brought a different but complementary perspective. Having run the agency responsible for protecting critical infrastructure, he understood how dual-use technology works—how the same innovation that solves a problem can create a new vulnerability. AI is perhaps the ultimate dual-use technology. It can accelerate medical research and accelerate bioweapon development. It can strengthen cybersecurity and enable cyberattacks. The same capability cuts both ways.
The regulatory question is where the real difficulty lies. How do you govern something that is advancing so quickly that yesterday's regulations are already obsolete? How do you protect national security without strangling innovation? How do you prevent bad actors from accessing dangerous capabilities without preventing good actors from doing good work? These are not rhetorical questions. They are the actual problem facing policymakers right now.
What made this conversation significant was not that it introduced a new fear—security experts have been warning about AI and biological weapons for years. What mattered was that two people with deep government experience, speaking on a major news platform, were saying clearly: this is real, this is now, and we need to figure out how to handle it. The debate over AI regulation is no longer abstract. It has a security dimension that cannot be ignored. The question is no longer whether to regulate, but how to do it in a way that actually works.
Citas Notables
The risk that AI could facilitate biological weapons development is not hypothetical but a genuine security concern— Ben Buchanan, Anthropic adviser and Johns Hopkins professor
La Conversación del Hearth Otra perspectiva de la historia
When Buchanan says the bioweapon risk is "not hypothetical," what does he actually mean? Is there evidence someone is already trying to do this?
He means the capability exists now. An AI system could provide information, design assistance, or optimization that would make developing a biological weapon easier than it was before. Whether someone is actively trying right now is a different question—but the fact that it's possible is the point.
So the problem isn't that AI is inherently dangerous. It's that AI makes dangerous things easier.
Exactly. A biological weapon is hard to make. It requires knowledge, resources, equipment. AI doesn't create the desire to make one, but it removes some of the friction. It's a force multiplier.
Krebs ran CISA. What does that experience tell him about how to actually regulate this?
He's seen how infrastructure gets attacked. He knows that regulation alone doesn't stop determined adversaries. But he also knows that making things harder matters. If you can slow someone down, you buy time. You create friction.
Is there a way to regulate AI without breaking it?
That's the question nobody has answered yet. You could restrict access to certain models. You could require safety testing. You could monitor for misuse. But each of those has tradeoffs. Restrict access too much and you slow legitimate research. Test too heavily and you slow innovation. Monitor too closely and you create surveillance concerns.
So we're stuck.
Not stuck. But we're in the early stage of figuring out how to govern something that's moving faster than our institutions usually move. That's the real challenge.