Bismarck is credited with the observation that laws are like sausages: it is better not to see them being made. The remark is probably apocryphal, which is itself somehow on-brand for a quote about the virtues of not looking too closely at things. Either way, the folk wisdom it encodes is real and ancient: what the prince doesn’t know, the prince can’t be held responsible for. Don’t tell me how the sausage gets made; just make sure the sausage gets made.
Like most folk wisdom, this is approximately right in the cases where it doesn’t matter and approximately wrong in the cases where it does. The specific way it goes wrong is the subject of this post.
In late January, shortly after Robert F. Kennedy Jr. was confirmed as Secretary of Health and Human Services, President Trump held a brief press availability and — asked whether Kennedy had the latitude to pursue his various heterodox health priorities — said, essentially: go wild. The phrase, more or less verbatim, was reported widely and treated by most commentators as either refreshing or alarming depending on their priors about Kennedy’s views on vaccines, fluoride, and the rest of the programme.1
What I want to point out is something slightly different: the press availability itself was the theoretically interesting move. Not because it authorized Kennedy to do things he otherwise couldn’t, but because it was public.
Here is the standard principal-agent story about willful ignorance. The prince tells the agent: I don’t want to know how you’re doing this. The agent acts freely, the prince maintains deniability, and if something goes wrong, the prince can plausibly claim he didn’t know. The logic is simple, the empirical referents are endless, and it is mostly right as a description of how discretion works in hierarchical organizations.2
I have argued before that willful ignorance can be not merely defensible but formally required — that even an unbiased politician will start conditioning decisions on information they’re known to possess, which means that keeping the principal genuinely in the dark is sometimes the only way to keep the decision process clean. That argument still holds. This post is about the specific way it doesn’t.
But notice what the story requires: the ignorance has to be real. Or at least, it has to be performed convincingly to the right audience. The prince who announces “I gave him total freedom” in front of a camera and a press corps has done something categorically different from the prince who quietly tells his agent that the normal rules don’t apply. He hasn’t just authorized the agent. He has enrolled himself — in front of witnesses — as a named party in whatever the agent subsequently does.3
There’s a legal concept that captures this precisely. Joint and several liability, in civil law, holds that when multiple parties are responsible for an outcome, any one of them can be held liable for the full amount — and they sort out the apportionment among themselves afterward. It is, in its way, an admission that the legal system sometimes cannot determine who did what, and so it assigns liability to the group and lets the group figure it out. The prince who publicly co-signs his agent’s freedom has voluntarily joined the group.
There’s a famous observation — variously attributed to Keynes, Getty, and approximately half of the financial journalists who have ever lived — that goes: if you owe the bank a hundred dollars, you have a problem; if you owe the bank a hundred million dollars, the bank has a problem. The intuition is clean: at sufficient scale, liability transfers from the debtor to the creditor. The creditor cannot afford to call in the debt without triggering consequences that harm the creditor more than the default would.
I want to suggest a variation on this theorem that applies to the prince-agent case:
If you quietly tell your agent to go wild, that’s your agent’s problem. If you announce it at a press conference, the bank has a problem.
The mechanism is different from the financial case. It’s not about scale; it’s about visibility. The public announcement doesn’t transfer liability because the agent becomes too big to fail. It transfers liability because the prince has created a record — observed by the meta-principal, meaning the public, the press, the opposing party, the Congress — that is inconsistent with subsequent claims of ignorance. The plausible deniability that willful ignorance is supposed to produce has been preemptively disclaimed by the very act of announcing it.
The Noem case is where this plays out in real time, and it is worth examining with some care.
Kristi Noem was removed as Secretary of Homeland Security on March 5th, following two days of bipartisan grilling on Capitol Hill over the administration’s immigration enforcement operations, including a Minneapolis operation in which two protesters were killed. The proximate cause of her removal was described, variously, as “drama” that “overshadowed the administration’s extremely popular immigration agenda” (per an administration official speaking to reporters), and as a mutual parting in which Noem would be “moving to be Special Envoy for The Shield of the Americas” — a position that, as of the announcement, did not exist and whose responsibilities were to be described at a future event in Doral, Florida.4
Notice, first, what Noem’s removal was not: a firing. She was relocated to a made-up ambassadorship in the Western Hemisphere, with Trump praising her as someone who “has served us well, and has had numerous and spectacular results (especially on the Border!).” The performance of no-fault separation is fully intact. This is not merely face-saving vanity; it is specifically the face that needs saving given the history. When Trump nominated Matt Gaetz as Attorney General and the nomination collapsed in public, he absorbed real political cost. The invented special envoy role is the technology that lets him avoid the “I appointed someone disqualifying” frame — or at least contest it.
What makes the Noem case theoretically interesting, though, is not the relocation. It’s the ad campaign. At her Senate Judiciary Committee testimony, Noem stated that the president had approved a $220 million campaign urging undocumented immigrants to self-deport — ads that featured Noem on horseback and were conducted primarily in English. The White House responded, immediately and on the record: “POTUS did not sign off on a $220 MILLION dollar ad campaign. Absolutely not.”
The president who said “go wild” is now, in front of Congress and the press, retroactively claiming he didn’t know about the expensive thing his agent did while going wild. This is not a surprising political move. It is, however, the exact moment at which the joint and several liability enrolled by the public “go wild” announcement comes due. The meta-principal observed the agent’s action, observed the prince’s prior public authorization, and is now watching the prince disclaim a liability he voluntarily created. Whether the disclaimer succeeds is an empirical question about what the meta-principal will tolerate. Whether it is formally coherent is not.
There’s a harder point lurking here that I want to gesture at without fully developing — partly because the development requires more space than this post has, and partly because I’ve written about it elsewhere.5
The legal system has a version of this problem, and its solution is quietly revealing. Joint and several liability holds that when multiple parties are responsible for an outcome, courts can hold any of them liable for the full amount — and this is true even when it is known that exactly one defendant actually committed the act, while the other is being held liable on a different theory (conspiracy, negligence, respondeat superior). The system isn’t making a mistake. It is operating as designed: rule systems cannot always resolve factual underdetermination, so they substitute liability assignment for truth-finding. The impossibility is conserved; it just gets moved around.
The structure is the same as the one I’ve been describing. The prince’s “go wild” announcement and the subsequent retroactive disclaimer jointly produce a result — Noem’s removal, Trump’s maintained deniability — that is formally inconsistent with the known facts. The system isn’t malfunctioning. That’s what systems that can’t fully police their own rules do: they generate outcomes that are locally coherent and globally inconsistent, and they rely on the meta-principal not looking too carefully at the seams.6
Which brings me to the sage advice I would offer, if asked, to Secretary Kennedy.
Noem’s replacement with a made-up Western Hemisphere envoy role is not a cautionary tale about being too aggressive or too visible or too willing to be the public face of unpopular operations. It’s a cautionary tale about what happens when the agent absorbs liability that the prince subsequently needs to disclaim. She went wild — as instructed, publicly, with the prince’s endorsement on record. When the meta-principal observed the results, the prince had two options: absorb the liability he had voluntarily enrolled in, or perform ignorance he had already precluded by announcing the arrangement in the first place. He chose the latter. The result is Noem on horseback, now apparently in charge of something called the Shield of the Americas.
The theoretical lesson isn’t that “go wild” is bad advice. It’s that “go wild, announced publicly, in front of the people who will later observe what you did” is a specific equilibrium with specific consequences for the agent. Noem’s mistake wasn’t going wild. It was becoming the visible face of the wildness. The prince’s public endorsement is not a shield; it’s a targeting reticle.
If you owe the bank a hundred dollars, you have a problem. If the prince announces at a press conference that you owe the bank a hundred million dollars, and the bank is Congress, and Congress is watching: the prince has a problem. And so, it turns out, do you.
With that, I leave you with this.
1 I am not, in this post, taking any position on the substantive merits of Kennedy’s health policy views, about which I have written before. I was more diplomatic then. The formal point applies regardless of where you come out on fluoride.
2 The formal treatment of this is in the principal-agent literature going back at least to the seminal work of Grossman and Hart (1983), and more recent work has examined how optimal delegation structures vary with the agent’s type and the observability of effort. I have a dog in this fight, so I’ll spare you the citation parade and simply note that the bilateral version of the model is well-understood. The three-player version — prince, agent, meta-principal — is where things get interesting.
3 There is a related point worth flagging. A public “go wild” announcement is also a kind of permission structure — an attempt to establish the agent’s faithful-type credentials before the meta-principal. I argued in 2013 that permission structures are almost always self-defeating, because a biased type has at least as strong an incentive to create one as a faithful type. “Never trust a man who says, trust me.” The present post is, in a sense, the same argument from the principal’s side: never trust a prince who says, I gave him total freedom.
4 The “Shield of the Americas” initiative was to be revealed at Trump’s Doral property. Readers who find the geography of this announcement interesting are invited to sit with that observation.
5 The relevant post is Can a Game Know Its Own Rules?, which develops the general argument that no rule system is fully self-enforcing, and connects it to Gödelian incompleteness and the conservation of impossibility. The current post is, in a sense, an application of that argument to the specific case of willful ignorance as a governance technology.
6 The criminal law version is particularly stark: courts can and do produce convictions that are jointly inconsistent given the known facts, and the system treats this as acceptable because the alternatives — acquitting everyone when guilt is uncertain, or requiring metaphysical certainty before any conviction — are worse. This connection between legal incompleteness and formal incompleteness is something I find genuinely fascinating and intend to return to.