Approve a Bad Thing, Career Over; Block a Good Thing, Nobody Notices

Picture the payoff matrix a drug reviewer actually faces. Approve something that later hurts people, and the failure has a name, a face, and a hearing room. Block — or just slow-walk — something that would have saved lives, and nothing happens to you at all, because the people it would have saved never knew the drug existed. One quadrant ends careers. The other is invisible. Put a rational person in that matrix and they will, on the margin, say no. Not because they’re cowardly or captured, but because the incentives were built that way.

That asymmetry is the whole game, and it’s worth taking seriously if you build anything that has to pass through a gatekeeper — which, increasingly, is everything. The framing surfaced in Naval’s “The AI Industrial Revolution” episode, in a stretch on why healthcare seems frozen while software eats the world: “If you approve a bad thing, your career is over. If you block a good thing, nobody notices.” That’s the seed. The rest of this is the part where we check whether it survives contact with the evidence.

The asymmetry has a formal name #

It does, and it predates the meme by decades. Economists call it asymmetric accountability: the situation that arises “when the consequences of one type of error are visible, attributable, and (potentially) career-ending, while the consequences of the other are diffuse, invisible, and nobody’s fault in particular” (Econlib). That’s the mechanism, stated cleanly. A Type I error (approve the harmful thing) lands on an identifiable victim. A Type II error (block the beneficial thing) lands on a statistical abstraction — the patients who quietly don’t get better. Only one of those generates a phone call from a journalist.

This isn’t theory retrofitted onto behavior. It was documented inside the FDA over fifty years ago. Former Commissioner Alexander Schmidt put it on the record: “The times when [congressional] hearings have been held to criticize our approval of new drugs have been so frequent that we aren’t able to count them” — while no comparable pressure ever materialized for a drug delayed or denied (Econlib, Drug Lag). The regulator wasn’t describing a personal failing. He was describing the gradient he was standing on. Inaction is the self-protective move, and the system rewards it.

What the bias costs #

Here’s where the engineer in you should perk up, because the cost is quantifiable and large.

The foundational work is Sam Peltzman’s, in the early 1970s. Studying the 1962 Kefauver-Harris efficacy amendments, he found that the number of new drugs approved fell “from an average of forty-three annually in the decade before the amendments to just sixteen annually in the ten years afterward” (Econlib, Drug Lag). Roughly a two-thirds collapse in throughput. Peltzman’s broader argument was that the foregone innovation — drugs that would have helped people but now simply didn’t ship — swamped the safety gains. (You’ll see his study dated 1973 in some retellings and 1974 in others; the figure is what matters, and the figure is 43 to 16.)

Casey Mulligan dragged the math into the present in a 2021 NBER paper, Peltzman Revisited (NBER w29574). The headline result is brutal in scale: even an accelerated approval process carried opportunity costs of roughly $1 trillion for a half-year delay on the COVID-19 vaccine in the US alone — ongoing infections, hospitalizations, and the wider economic seizure that every extra week bought (Becker Friedman Institute summary). That’s not the cost of a slow process. That’s the cost of a fast one.

And the human version is sharper still. Alex Tabarrok tallied the gap between Pfizer’s Emergency Use Authorization application on November 20, 2020 and authorization on December 11 — twenty-one days — during which roughly 14,696 Americans died of COVID-19, many of whom could have received doses that were already manufactured and sitting in warehouses (Marginal Revolution). The FDA wasn’t reviewing a mystery compound; it had the full Phase 3 data in hand. The delay was procedure, committee scheduling, and an unwillingness to outrun the standard process — because outrunning it created visible accountability if anything went wrong, while the deaths from waiting stayed invisible. The asymmetry, compressed into a single measurable three-week window. That’s the thesis as a number you can put on a slide.

The honest counter-section #

If the story were this tidy, “abolish the gatekeeper” would be the obvious conclusion, and it isn’t. The strongest rebuttals deserve full weight.

The precautionary side genuinely works sometimes. The founding case is thalidomide. FDA reviewer Frances Oldham Kelsey refused to clear the sedative for the US market, requesting more safety data from the manufacturer roughly every sixty days from 1960 until the application was withdrawn in March 1962 (Wikipedia: Frances Oldham Kelsey). Thalidomide had already been approved in Canada and more than twenty other countries, where it caused severe birth defects on a scale measured in the thousands; her holdout is credited with sparing the US that catastrophe, and she received the President’s Award for Distinguished Federal Civilian Service in 1962. That is not a just-so story — it’s a real save. The “no” reflex that the asymmetry encourages is, in this one famous instance, exactly what you wanted. Any honest account has to hold both: the same caution that produced the invisible graveyard also produced Kelsey.

The system fails in the other direction too, and that’s revealing. OxyContin is the usual proof that the FDA approves bad things. The 1995 approval involved a reviewer, Curtis Wright, who met privately with Purdue Pharma, helped shape the safety review, signed off on label language calling addiction potential “very rare” without clinical evidence to back it, and then left the agency within two years for a far better-paid job at Purdue (GovFacts). Congress held no hearings while prescriptions quadrupled. But look at the actual failure mode: it wasn’t excess caution, it was the revolving door. When a regulator expects to join the industry he regulates, the asymmetry inverts — approval becomes the career-protecting move. The deeper lesson isn’t “too much caution” or “too little.” It’s that accountability tracks careers, not outcomes, and a career can be bent in either direction.

The bias isn’t immutable — incentives can be redesigned. The 1992 Prescription Drug User Fee Act let industry fund FDA reviews in exchange for binding timelines, and median approval times fell sharply over the following decade. The precise figures get garbled in circulation; a fair reading is that medians dropped from roughly thirty months toward something under twenty months for standard reviews, with priority pathways faster still (Wikipedia: PDUFA). The magnitudes are debated and worth pinning to a primary source before you quote a hard number — but the direction is the point. Deadlines, fee structures, and accountability metrics moved a “no by default” institution measurably toward “yes, faster.” The gradient is not destiny; it’s a parameter.

And the regulator may not even be the binding constraint. A competing explanation for drug lag isn’t reviewer timidity at all — it’s clinical-trial cost, patent clocks, and sponsor economics. A large share of the “missing” drugs are ones no sponsor ever attempts, because the cost-benefit doesn’t pencil out long before anyone reaches a reviewer’s desk. That’s a market failure upstream of the gate, and blaming the gatekeeper for it is a category error.

The thing to watch #

If the asymmetry thesis is right, the cleanest test would be to gut approval capacity and see what breaks. We’re getting a messy, unplanned version of that experiment right now. The 2025 federal workforce reductions hit HHS hard — the contemporaneous reporting confirms “up to 5,000 staff across HHS may be dismissed” and “at least 230, perhaps more” in the FDA’s device office specifically, with the department declining to detail exact figures (BioPharma Dive). Larger FDA-specific totals in the low thousands, partial rehiring efforts, vanished label databases, and reviewers recruited as volunteers have all been reported elsewhere, but those specifics outrun what that single source establishes, so treat them as unconfirmed until a primary tally lands.

Here’s the irony the asymmetry predicts. Cutting reviewers doesn’t make the gate say “yes” — there’s no one left to say it. It makes the gate say nothing, which defaults to no. Removing the bottleneck without redesigning the incentive doesn’t deliver abundance; it delivers a slower, quieter graveyard. The Peltzman cost compounds while the headcount falls.

The builder’s takeaway isn’t “regulation bad.” It’s that every gate you ship code, drugs, or hardware through has a payoff matrix attached to the human on the other side, and that matrix is almost always tilted toward inaction — because someone has to own the visible failure and no one owns the invisible one. If you want a gatekeeper to move faster, don’t yell at the gatekeeper. Change which error costs them their job. That’s the only knob that has ever worked.