Almost overnight, AI has gone from a chatbot that could give advice for your sore throat to the most important tool in modern medicine.
Human doctors remain the decision makers in your medical treatment, but AI has an astoundingly large role. Drug companies use AI-designed molecules to accelerate vaccine development. AI algorithms are embedded into pacemakers and insulin pumps. Brain implants use AI to decode neural signals. AI is a trusted second pair of eyes for radiologists and pathologists reviewing medical images, helping to diagnose tumors and spot issues humans miss. AI chatbots triage patient symptoms, and suggest possible diagnoses.
The list is long and mind blowing. It wasn't so long ago AI didn't exist. Now its helping to edit genes.
But there are lines we haven't crossed. Yes, AI helps analyze data and design treatments. But an AI-designed component has never had the ability to be injected into a human and attack live cells -- even deciding to kill them.
Until now.
Researchers from the Technical University of Denmark have created AI-designed T cells that enter the blood of live humans and kill cancer cells.
Normally, your immune system sends T cells to attack when there are foreign cells detected in your body. The diabolical thing about cancer is, they are your own cells. They are not foreign, so the natural alarms don't go off. It's a disease hiding in plain sight. The deadly cancer cells do have unique marks -- unusual proteins on their surface (MHC molecules). But all too often, the T cells don't have the right receptors to see these marks. The cancer is left alone to grow. The results are almost always deadly.
In the study published this year, researchers explain how they used AI to design protein structures that can "see" MHC molecules - those unique marks on cancer cells. And they engineered T cells in a lab that had these MHC detectors incorporated. The ultimate goal is to infuse these T cells into the blood stream of a live human and let them do their thing: kill cancer. But first, they did it in a test tube.
And it worked.
The AI-engineered cells were able to recognize cancer cells, release toxic proteins, and kill them. In this case it was melanoma. But the AI-designed protein-keys that reprogram a patient's T cells could be designed to target any cancer, with unprecedented accuracy. In fact, as part of the same study, the researchers did a second test to prove they could create MHC detectors for other cancers. And they were able to us AI to generate binders (what would attach to T cells and make then "see" cancer) that recognized the markers of an individual patient's unique tumor. Not only did they engineer T cells that kill cancer, they proved the method's potential to target any cancer.
It's truly amazing. A milestone in medicine. When this approach makes it into real-world, clinical practice, it will be the first time something created by AI has been placed into the human body for the purpose of killing human cells.
But if T cells can be engineered to target any cancer, could they also be engineered to kill any cells? This potential makes the study a milestone in bioethics.
What if, instead of targeting cancer cells, AI-engineered T cells were designed to attack healthy tissues? Imagine this new weapon somehow used on you. If you went to a doctor and complained something inside you was killing your own cells, they would find no evidence you were tampered with. After all, there's no poison in your blood, just T cells attacking other cells. Your symptoms would look like an autoimmune disease. They'd tell you it looks like lupus, or multiple sclerosis or myasthenia graves -- or any number of vicious but known diseases where your immune system goes haywire. Then the AI engineered cells would continue their work until you faced multi-organ failure and it was too late.
It's comforting that it's not so easy to get those evil T cells into your body. The process of receiving AI-engineered cells would likely involve a formal blood infusion. You'd have to sit in a chair and be hooked to an IV drip and hang out for some amount of time. In other words, no one could "secretly" give you the treatment. You'd know if it was happening.
If some evil-minded entity had the scientific know-how to manufacture this horrific bioweapon, they would face the challenge of how to get it in our bloodstream.
That's where the bad news comes in. While an IV infusion is currently the standard for T cell therapies, it is possible to introduce weaponized T cells in less obvious ways. Like a small-volume injection directly into the bloodstream, possibly disguised as a vitamin shot or other routine medication. No one would ever know.
Another possibility: modified stem cells or encapsulated delivery vesicles could slowly release AI-programmed T cells over time via a skin implant. Done on a single person, it becomes a diabolical method of assassination. Done on a larger scale, it could wreak chaos and we'd never know the source.
Obviously, these fears are not imminent or even feasible today. The amazing work of the Danish researchers should be advanced as fast as possible. There are real possibilities to end cancer in ways we've never conceived of before.
But we've crossed a line in creating AI-engineered cells that can act on the human body. Perhaps there should be safeguards that mitigate biological threats. The Trump Administration's recent "AI Action Plan" specifically makes sure AI can not be "woke". But it is silent on AI being misused as a bioweapon.
Isaac Asimov suggested three rules to program into robots to safeguard humanity, in case they decided to get rid of us. The first was that a robot may not injure a human being. Might be time to make sure any AI model has that rule built in somehow. Before any more lines are crossed.