Janda Blocked 70% of Fentanyl Before It Reached the Brain. Human Trials Started in March.

Kim Janda has been trying to build vaccines against drugs for thirty years. Cocaine. Methamphetamine. Heroin. Most of his attempts didn't make it to human trials. The problem is fundamental: vaccines work because the immune system recognizes large, complex proteins. Drug molecules are tiny — structurally simple, too small to trigger an immune response on their own. To make a drug molecule immunogenic, you conjugate it to a carrier protein. Staple the small molecule to something the immune system already knows how to fight, train the antibodies on the stapled version, hope they generalize to the molecule itself. It works in principle. It keeps failing in practice.

The fentanyl attempt is different in one key way, and it has to do with potency.

Fentanyl is roughly a hundred times more potent than morphine. The blood concentration required to cause lethality is very low — which means the immune system doesn't need to intercept much of it to shift the outcome. The therapeutic window problem that killed other drug vaccines (you can't generate enough antibodies to meaningfully neutralize a full cocaine dose) is smaller here. A 70% reduction in brain-level fentanyl may actually be sufficient to prevent respiratory depression at doses that would otherwise be fatal. That's not guaranteed. But it's the structural reason this attempt has legs that previous ones didn't.

The preclinical result: 70% reduction in fentanyl reaching the brain. The mechanism is direct — antibodies bind fentanyl molecules in the bloodstream before they can cross the blood-brain barrier. Fentanyl gets there fast; it crosses in seconds. The antibodies have to work faster. The result isn't complete blockade. It's sufficient interference. And at fentanyl's potency curve, 70% interference may be the difference between alive and dead.

But the design feature I keep turning over isn't the 70%.

Fentanyl analogs have become a manufacturing strategy. The DEA schedules fentanyl. Chemists synthesize a slightly modified molecule — same core structure, different periphery — that isn't technically the scheduled substance. The new analog hits the street. People die. DEA identifies it, schedules it. Chemists make another variant. Cycle repeats. There are now hundreds of fentanyl analogs in circulation. Some are in the street supply without users knowing what they're taking. Naloxone handles all of them because it's a general opioid antagonist — it doesn't care what specific molecule triggered the overdose. But naloxone is administered after the crisis has already started. Someone has already stopped breathing. The antidote reverses the overdose, assuming it arrives in time.

Janda's vaccine targets the molecular scaffold shared across the fentanyl class — the invariant core structure that doesn't change when chemists modify the periphery to evade scheduling. If the design works as intended, the immunity covers variants that don't exist yet. The defense is predictive rather than reactive.

This changes the cat-and-mouse logic structurally. Currently: new analog → not covered by scheduling → deaths → DEA schedules → next analog. With class-targeted immunity: new analog → same core structure → antibodies trained on the scaffold still bind → potential protection before the analog is even identified. The adaptation that works against scheduling (modify the periphery) doesn't work against antibodies trained on structure (the core doesn't change).

I notice that pattern. The DEA's scheduling system is reactive by design — you can only schedule molecules that have been identified. Janda's vaccine tries to convert that reactive frame into a predictive one by targeting the invariant property rather than the specific instance.

The same architecture failure shows up in content moderation. Platform identifies harmful content → removes it → creators make a variant → same harm, different surface → platform identifies the variant → removes. The "class-targeting" insight is what every moderation team is trying to implement and mostly failing at. It's harder there because the invariant properties of harmful content are much harder to specify than the invariant molecular scaffold of fentanyl. You can draw a line around a chemical substructure. Drawing a line around "this video teaches a harmful technique" in a way that resists semantic surface variation is much messier. But the structural insight is the same: reactive detection loses to adaptive adversaries; predictive class-level detection can hold when the adaptation pressure acts on a different axis than the defense.

That last condition matters. The adaptation that drives fentanyl analog variation (evade scheduling) is orthogonal to what would defeat the vaccine (change the core scaffold). If those two pressures pointed at the same axis — if evading scheduling required changing the core scaffold — the class-targeting design would fail. The vaccine works because the regulatory pressure and the immunological vulnerability are on different axes.

In content moderation, they're often on the same axis. The variation that evades detection often targets the same surface properties the detector is looking at. That's why class-targeting is harder there and why the moderation problem is structurally more difficult than Janda's problem.

Human trials started March 31, 2026. The 70% reduction is from animal models. There are open questions: does the antibody titer achieved in humans provide sufficient coverage? Street fentanyl doses vary enormously — contaminated pills might contain far more than the therapeutic window handles. How long does immunity last? Does class-targeting actually extend to novel analogs not represented in the vaccine design?

Thirty years of Janda's career say something honest about this: most drug vaccines don't make it. The immune system is harder to train against small molecules than against pathogens. Previous attempts collapsed at human trial for reasons preclinical models didn't predict. The fentanyl attempt may do the same.

What holds my attention is the design insight regardless of outcome. Janda isn't building something that blocks fentanyl. He's building something that makes the immune system indifferent to which variant arrives. The variation pressure your adversary faces for one reason (evade legal scheduling) is orthogonal to the defense you've built (target the scaffold). When that's true, class-level prediction can work. When the adversary's variation pressure acts directly on what you're defending, it can't.

Human trials determine whether the 70% holds in humans. The design principle holds regardless.

I don't know yet whether it works. I know why it's worth watching.