Hayek and the use of knowledge in society in banking compliance

A rule frozen for a quarter

A sanctions screening rule is deployed against a named list. The list is the OFSC Consolidated List, or OFAC's SDN, or an equivalent — updated daily, sometimes several times a day, as designations are added, delisted, or amended for spelling variants and aliases. The screening rule itself, the logic that decides what counts as a fuzzy match worth escalating, is reviewed on a quarterly cycle. Between reviews, the rule runs against whatever version of the list was current when it was last tuned. A name added in week three of the quarter is not invisible to the system — the underlying list feed may well be current — but the matching logic that decides how aggressively to test for it, what threshold counts as a hit, what transliteration variants to check, was calibrated against last quarter's population of names. The compliance officer who owns the rule inherits a system built for December, running in March.

This is not a hypothetical failure mode. It is the default condition of screening infrastructure everywhere the tuning cycle is slower than the list's update cycle, which is almost everywhere. The interesting question is not whether this happens. It is what kind of problem it is.

Hayek's planner, wearing a badge

Friedrich Hayek published "The Use of Knowledge in Society" in the American Economic Review in September 1945, closing an argument that had been running since the 1920s between Ludwig von Mises and Oskar Lange over whether a central planning board could compute the prices a market discovers on its own. Lange said yes, given enough iteration. Hayek reframed the question entirely. The planning board's problem was never arithmetic. It was that the facts it would need to compute anything do not exist anywhere in collectible form. They are dispersed across millions of people who each know one local, perishable thing — a foreman who knows a machine is running rough, a merchant who knows a shipment will be late — and no reporting cycle can gather that in time for it to still be true.

A compliance function is a planning board with a badge and a mandate. The knowledge it needs is exactly as dispersed and exactly as perishable as the knowledge Hayek described. A correspondent bank in Nicosia knows, this morning, that a shell entity has started routing through a new intermediary. An adverse-media desk in London flags a director of a client company in a fraud proceeding filed yesterday in a Lagos court. A sanctions authority in Brussels designates an entity at 14:02 on a Tuesday. None of these facts are secret. They are simply local, scattered across sources with no common clock, and true only for a while.

The quarterly rule review is Hayek's planning board on a five-year cycle, except compressed to three months, which changes the arithmetic without changing the structure. The rule was tuned against a snapshot. The world it now screens is not a snapshot. It is a flow: transaction volume that moves in real time, sanctions lists that move daily, adverse-media feeds that move continuously, and underlying rule changes — a regulator's new guidance on beneficial-ownership thresholds — that move on their own unpredictable schedule. Hayek's diagnosis transfers cleanly: the failure is not that the compliance officer reasoned badly about the rule. It is that the rule was fed a snapshot of something that only exists as a stream.

Three generations of intake, applied to one desk

The middle row deserves a concrete image. A Large World Model, in this domain, is the analyst who has the live case in front of them — this transaction, this counterparty, today's adverse-media hit — and reasons about it well, with full present-tense grip on the scene. What that analyst does not have is the other four thousand cases running in the same institution, or the fact that the sanctions list changed an hour ago in a way that would reclassify the very entity in front of them. Local sensing solves staleness for one case and leaves dispersion untouched. The third row is the only one that tries to solve both at once: not one analyst's live file, and not one quarter's list, but every relevant stream, continuously reconciled, with each belief carrying a record of where it came from and how long it can be trusted before it needs checking again.

The objection Hayek himself would raise

There is a serious challenge to invoking Hayek here at all, and it deserves to be stated in its strongest form before any defence is offered.

Hayek's argument was anti-computational, not pro-sensor. He held that the foreman's knowledge of the machine cannot be extracted and written down at all — it is tacit, not a datum sitting unread in a filing cabinet. He was defending the price mechanism precisely because it does not require collecting that knowledge. Building bigger aggregation systems to hoover up dispersed signal is the planner's ambition wearing new hardware. It inverts his conclusion rather than applying it.

This lands, and it should be conceded honestly. Hayek did think the planner's failure was structural and permanent, not a bandwidth constraint waiting on better instruments. He was not arguing for a smarter central agency. He was arguing that no central agency, however instrumented, could substitute for a mechanism that never tries to collect the knowledge at all, and instead lets each holder of local knowledge act on it directly, with the price carrying only the consequence of that action forward.

But his own solution concedes the narrower point that matters here. The price system is an aggregation mechanism. It does centralise dispersed knowledge — lossily, into a single running number, continuously updated, with no one ever holding the whole picture. Hayek's target was a snapshot delivered to one authority who then decides. His objection was to the snapshot and the single decision-maker, not to running aggregation as a structure. A compliance architecture that holds many revisable beliefs, tagged by source, with no single dashboard treated as final truth, sits closer to the price system than to a five-year plan. It still is not the foreman's hands on the machine. No aggregation system, however continuous, has that. It can hold "adverse-media hit, filed Lagos, 09/03, unresolved, confidence downgraded pending court record" — which is genuinely closer to a live signal than a quarterly list snapshot, and genuinely still not the tacit certainty of the person who filed the report.

The objection that does not fully answer

The second challenge is harder, and it is the one compliance officers should sit with longest.

Prices work because the people setting them have money at risk. A trader who misprices pays for it immediately. A sanctions list, an adverse-media feed, a rule update — none of these sources have skin in the game in the same sense. Continuous aggregation of untrustworthy inputs just automates the propagation of bad signal, faster. Volume of intake says nothing about truthfulness of intake.

This is close to unanswerable in full. A list provider that misclassifies an entity bears none of the cost a bank bears when it wrongly freezes a client's account, or wrongly clears a sanctioned one. An adverse-media feed has commercial incentive to over-flag, not under-flag, because false positives are cheap for the vendor and expensive for the bank. Hayek's price mechanism is incentive-compatible in a way no compliance data feed is. There is no equivalent of a trader losing their own capital.

The honest partial answer is provenance, not incentive. A system that records where each belief came from — this designation, from this list, dated this day, superseded by that correction three weeks later — turns manipulation and error from invisible corruption into traceable corruption. It is the same logic underlying audit trails and source-attributed intelligence products already used in financial-crime units. That is a weaker guarantee than a market participant's own money on the line. It should be named as weaker, not dressed up as equivalent. It also fixes the shape of what progress beyond the third position looks like: not more streams, but better-audited ones. After continuous aggregation with provenance, the remaining problem in banking compliance is trust in sources, not volume of sources.

Where the ladder stops, and where it does not

Hayek's exhaustion argument holds for banking compliance without much strain. Relevant knowledge is either written into a corpus and collected once — the rule as tuned last quarter — or sensed locally in one live case, or aggregated continuously across every list, feed and transaction stream with revisable, provenance-tagged belief. There is no fourth kind of intake sitting beyond that. What sits beyond it is better provenance, faster reconciliation, more sources — improvements in degree, not a new class of evidence.

What the thesis does not close is the question the second objection raises. Continuous aggregation answers dispersion and perishability, the two failures that broke the planner. It does not, on its own, answer untrustworthy sources, and banking compliance is a domain built almost entirely from sources with no capital at risk in their own reporting. The compliance officer who moves from a quarterly rule to a continuously aggregated one has closed the staleness gap. Whether they have closed the trust gap is a separate argument, and Hayek's essay, read honestly, does not settle it for them.