The Last Audit: What We Actually Know About the World's Official Gold

Start with a number that everyone in the gold market uses and almost nobody can prove: 36,000 tonnes. That is roughly what the world's governments and central banks say they hold — about a fifth of all the gold ever mined, worth at current prices in the region of three trillion dollars. The figure appears in every annual demand report, every central bank speech, every chart of official-sector buying. It is the load-bearing statistic of the gold market. And if you ask what physically stands behind it, you discover something that surprises most people the first time they hear it: the number is, in essence, a collection of assertions. Reported by the holders, compiled by international institutions, repeated by analysts, and verified — bar by bar, against a current list, by an independent party — almost nowhere.

This is not an article claiming the gold is gone. The overwhelming likelihood is that the world's official gold is exactly where its owners say it is; the institutions involved are old, conservative and obsessively procedural, and a missing-tonnage scandal at a major central bank would be one of the largest financial crimes in history. But there is a difference between gold that is there and gold that is known to be there, and the distance between those two states is the subject of this piece. It is a distance measured in audit reports that were last refreshed when Eisenhower was president, in compartment seals inspected instead of bars, in foreign custodians whose word substitutes for an owner's eyes, and in reporting conventions that allow a tonne out on loan to count the same as a tonne on the shelf.

To understand the audit gap you first have to understand what a reported reserve figure actually is, because it is not one thing. It is a composite of at least three layers, each with its own standard of proof, and the layers are rarely distinguished in public discussion.

Three kinds of 'held'

The first layer is gold in a vault on home soil, under the owner's direct control. The Bundesbank's Frankfurt vaults, the Banque de France's La Souterraine, the US Mint's facility at Fort Knox. This is the layer the public imagines when it hears a reserve figure, and it is the only layer where physical verification by the owner is even straightforwardly possible.

The second layer is gold held abroad at a foreign custodian — overwhelmingly the Bank of England in London and the Federal Reserve Bank of New York. Dozens of countries keep metal in these two vaults. The practice is not sinister and not new: London and New York are the two places where official gold actually trades, and metal already sitting in the settlement vault can be sold, swapped or transferred with a ledger entry instead of an armoured convoy. Much of it is a fossil of the Bretton Woods system, when the dollar was convertible into gold and foreign central banks accumulated metal in New York as a matter of routine. But custody changes the evidentiary standard. A German bar in New York is not inspected by Germany on a schedule; Germany receives statements from the Federal Reserve. The owner holds a very credible piece of paper.

The third layer is the least discussed: gold that is reported as held but is not, at that moment, in anyone's vault on the owner's behalf — because it is out on lease or pledged in a swap. Central banks lend gold to bullion banks to earn a yield; bullion banks sell it and deliver later; the gold finances refineries, jewellery manufacturers and miners' hedging programmes in between. IMF reporting guidance has historically permitted gold out on loan to remain inside the headline reserve number, on the logic that it remains an asset of the central bank. Which it does — in the same way that money you have lent a friend remains your asset. The asset is real. It is just a claim, not a bar.

What a physical audit actually involves

It is worth being precise about what the word 'audit' means in this context, because casual usage has blurred an exacting technical exercise into a synonym for any official reassurance. A genuine physical audit of a sovereign gold holding is closer to a forensic inventory than to a bank's annual accounts review, and it proceeds in a fixed sequence that has changed little since the mid-twentieth century, even as the tools used within each step have improved.

Reconciliation against the bar list

Every properly documented holding is described by a bar list: a ledger recording, for each bar, a serial number, refiner mark, gross weight, fine weight and assay stamp. The first stage of an audit is not touching a single bar — it is reconciling the current bar list against the previous one, bar by bar, to identify additions, removals and any serial number that appears without a matching transaction record. Discrepancies at this desk-based stage, before anyone opens a vault door, are usually clerical; but a genuine audit treats every unexplained line as a flag requiring physical resolution, not administrative closure.

Physical count and weighing

The second stage is where the exercise becomes expensive. Bars are removed from their racks, individually weighed on calibrated scales accurate to a fraction of a gram, and checked against the ledger entry for that serial number. For a holding the size of the US reserve — more than four hundred thousand individual bars across several depositories — a full count is a multi-year undertaking requiring dedicated staff, secure logistics for moving bars within a vault, and a shutdown of normal operations in the sections being counted.

Statistical sampling and assay

Because drilling every bar to test its interior is neither necessary nor desirable, auditors work from statistical sampling theory: a randomly selected subset, large enough to give a specified confidence level, is drilled or scanned. Traditional assay involves taking a drilled core and testing it chemically or by fire assay for fineness; modern facilities increasingly supplement this with ultrasonic and X-ray fluorescence scanning, which can detect a non-gold core — a tungsten-filled bar, for instance — without destroying the sample. Neither method, applied to a five per cent sample, guarantees that a single doctored bar elsewhere in the holding would be caught. Confidence, not certainty, is what a sampling regime buys, which is precisely why the Bundesbank's decision to melt and recast its entire repatriated holding, rather than sample it, sits in a different category of proof.

Chain-of-custody documentation

The final and least glamorous stage is paperwork: establishing that every bar counted today can be traced, transaction by transaction, back to the last point at which it was physically verified, with no gap in the record wide enough to hide a substitution. It is this stage that the US position struggles with most, because the chain running back to 1953 has documented interruptions — bars moved for assay, compartments reopened for the 1974 visit, records not always contemporaneously filed. None of the released documentation demonstrates a discrepancy. It does demonstrate that the chain, in the technical sense auditors use the term, is not unbroken.

The IMF's reporting standard, and what it does not require

The 36,000-tonne figure exists as a usable global number because of an international accounting convention, not because of an international verification regime. Members of the International Monetary Fund report their reserve assets — gold included — under the Special Data Dissemination Standard, a framework built primarily to promote timely, consistent disclosure of the categories and broad magnitude of reserves, not to certify that a physical count underlies any given entry.

Under the SDDS reserves template, monetary gold is reported as a single line, typically in both fine-troy-ounce or tonne terms and a market-value equivalent, alongside foreign-exchange holdings, IMF reserve positions and other reserve assets. The template asks a country to state what it holds; it does not ask a country to prove it, and it contains no field distinguishing gold held in a home vault, gold in a foreign vault, or gold currently out on loan. All three collapse into the same reported tonnage. This is a design choice with a defensible rationale — reserve reporting exists mainly to give markets and multilateral institutions an early warning of stress in a country's external position, not to serve as a bar-by-bar registry — but the consequence is that the IMF template, for all its rigour on timeliness and definitional consistency, is silent on the one question this article is asking.

The World Gold Council, whose published tables are the version of the 36,000-tonne figure most investors actually see, is transparent about its own position in this chain: it aggregates IMF-reported data and central bank disclosures, and explicitly does not conduct independent verification. Every publication in the chain, from national central bank to IMF to World Gold Council to the financial press, is reporting the same underlying self-declared number with an added layer of institutional credibility at each step — and at no step does anyone re-open the vault.

1953: the last time anyone counted

The United States reports 8,133.5 tonnes, the largest official holding in the world, and it is the case that tells you the most about how official gold verification actually works — because the American audit history is the best documented in the world, thanks to decades of freedom-of-information requests by researchers who refused to let the question drop.

The last full physical audit of the US gold stock was completed in 1953. Bars were counted, weighed and sampled across the depositories. It was a genuine audit in the modern sense, and it has never been repeated. What followed instead was a series of gestures, each smaller than the last. In 1974, amid conspiracy theories that Fort Knox had been emptied, the Treasury opened a single compartment to a delegation of congressmen and journalists. They walked around, held bars, posed for photographs. A few bars were assayed. It answered almost nothing — the vault has fifteen compartments — but the photographs did their work, and the visit entered folklore as 'the audit'.

From the mid-1970s onward, the practice settled into what the Treasury calls 'continuing audit' or 'seal verification'. After the 1974 visit, compartments were closed and sealed with official wax or tape seals bearing a committee's marks. Each year, auditors check that the seals are unbroken. If the seal is intact, the gold is presumed to be as it was when the compartment was last opened — in some cases, as it was verified to be decades ago. It is an audit of the door, not the room.

Documents released under freedom-of-information law have since shown that at least some compartments were opened after being sealed — for assay sampling, for movement between facilities, for the 1974 visit itself — and resealed under procedures that were not always fully documented. None of this demonstrates loss. It demonstrates something quieter and, for a statistician, almost as uncomfortable: the paper trail between the 1953 count and today's reported figure has gaps, and the institution's position is that the gaps do not matter because the institution says so.

The custodians: Threadneedle Street and Liberty Street

If Fort Knox is the famous vault, the important vaults are in London and New York. The Bank of England's basement holds gold for the UK and for around seventy other official holders — one of the largest concentrations of bullion on earth, none of it visible to the owners between statements. The Federal Reserve Bank of New York's vault, twenty-five metres below Liberty Street, holds metal for roughly three dozen foreign governments and international institutions. Between them they custody a substantial fraction of the world's reported official gold.

Both institutions are scrupulous operators with century-long records. The New York Fed's procedures are a marvel of paranoid craftsmanship: a triple-locking system in which no single person knows a full combination, a vault door that seals on a time lock, bars moved between numbered compartments so that even staff cannot map a country's full holding. But the central fact about custody is structural, and no craftsmanship changes it: the owner verifies by attestation. The Bank of England and the Fed issue statements and confirmations; auditors of the owning central banks rely on those confirmations. When you hear that a country's gold was 'audited' while sitting in London, what usually happened is that someone checked the custodian's letter against the owner's ledger.

For most owners, most of the time, this is entirely rational. Physical audits of tens of thousands of bars are expensive, slow and disruptive; the custodians are among the most credible institutions in finance; and the gold is in London or New York precisely because that is where it is useful. The vulnerability only becomes visible at the aggregate level: a very large share of the world's most-quoted gold number ultimately rests on the internal records of a handful of institutions, none of which opens its vault to outsiders for a full independent count.

London: allocated accounts, unallocated risk

The Bank of England distinguishes between allocated gold, where specific serialised bars are recorded against a single owner and cannot be lent, moved or used without that owner's instruction, and unallocated gold, a fungible credit balance representing a claim on a given weight of metal from a general pool. Most official reserve accounts at the Bank are held allocated, which is the more conservative arrangement — but allocation solves the ownership question, not the verification question. An allocated bar in the Bank's basement is still counted, weighed and confirmed by the Bank's own staff, on the Bank's own schedule, using the Bank's own records. The owning central bank receives a confirmation, not an invitation to watch the count.

New York: the sub-vault system

The Federal Reserve Bank of New York operates on a comparable principle with an additional layer of internal fragmentation: each depositing country or institution is assigned one or more numbered compartments, and bars are moved between compartments as transactions between account holders occur — a sale settled by relocating bars from one country's compartment to another's, rather than moving metal in or out of the building. It is an elegant, low-risk way to settle international gold transactions, and it is also the reason a foreign government's holding in New York is, in a very literal sense, defined by which shelf its bars currently sit on according to the Fed's internal records — a fact known with certainty only to the small number of Fed staff who manage compartment allocations.

What owners can and cannot ask for

Both custodians permit periodic visits by depositor representatives, and both will, on request, arrange for an owner's officials to witness the physical existence of bars in aggregate — a spot check rather than a full count. Neither custodian's standard arrangement gives a depositing central bank standing to demand a full independent recount of its own holding on its own timetable. That would require a bespoke agreement, and in the public record there is no example of one of the largest depositors — the United States, Germany, Italy, the IMF itself as a gold-holding institution — obtaining a full, current, third-party count of its custodied metal from either vault.

Germany did it properly

There is one case that shows what genuine verification looks like, and it is the Bundesbank's repatriation programme. In 2013, after years of public pressure — a citizens' campaign, parliamentary questions, a critical report from the German federal audit office noting that the Bundesbank had never physically verified its foreign-held gold — Germany announced it would bring home 674 tonnes: 300 from New York, 374 from Paris.

The operation finished in 2017, three years ahead of schedule. And the detail that matters is what the Bundesbank did with the bars when they arrived. Every returned bar was melted down and recast to current London Good Delivery standard — a process that weighs the metal, assays its fineness and produces a new, verified bar with a new serial number. You cannot hide a tungsten core through a melt. The Bundesbank reported no discrepancies. It also published a bar list: serial number, refiner, weight, fineness, for every one of its bars, at home and abroad.

Germany thus became the only major holder to have both repatriated at scale and publicly verified at scale in the modern era. The exercise also quietly demonstrated the point of this article: it took one of the world's most powerful central banks four years, specialist logistics and the melting of 674 tonnes of metal to convert a custodian's attestation into physical certainty for just over a fifth of its holding. Nobody else has done it. The cost and awkwardness of the exercise is precisely why.

The leasing layer, and the number nobody publishes

The third layer of the composite — gold out on lease or in swaps — is where transparency falls off a cliff. Central banks do not publish their outstanding gold loans. The Bank for International Settlements publishes aggregate swap positions in its reports, which gives analysts a partial window into one corner of the market, and the swings in that number are watched closely by the small community that tracks it. But the total volume of official gold currently out of official vaults — lent to bullion banks, sold onward, financing the working metal of the jewellery and refining trades — is an estimate, not a statistic.

The reporting convention makes this opaque by design. When a central bank lends gold, the gold typically stays in its reserve total, because the loan is still its asset. There is no asterisk in the headline number saying 'of which X tonnes are currently someone else's working inventory'. In the early 2000s, when the official sector was a heavy lender, analysts estimated the outstanding lease book in the thousands of tonnes. Estimates today are lower, but they are still estimates. The practical consequence is that the 36,000-tonne figure overstates, by an unknown amount, the metal that is physically in official custody and immediately deliverable by the owner without waiting for a loan to unwind.

Does that matter for price? At the margin, yes, in both directions. Lending adds effective supply to the market even while the gold stays in the reserve statistics — a quiet double-count that the market spent much of the 1990s arguing about. And a widespread demand for return — a lease-market squeeze, or a wave of repatriation requests at once — would force that paper layer back into physical form, with the bullion banks caught in the middle having to buy metal to return it. The market's memory of the 1999 Washington Agreement, which capped official sales and lending precisely because the overhang was destabilising, is one reason the official sector now lends far more cautiously.

What honest disclosure would look like

A commercial company that pledged a third of its inventory as collateral, or lent it to a counterparty, would be required under ordinary accounting standards to disclose the encumbrance in a note to its accounts — the asset stays on the balance sheet, but a reader can see that part of it is not immediately available. No equivalent disclosure attaches to official gold. A central bank's published reserve figure carries no note stating what fraction, if any, is currently out on loan, pledged in a swap, or otherwise unavailable for immediate delivery. Some central banks disclose gross lending income in their annual accounts, which allows an informed outsider to infer that lending is occurring and roughly at what scale, but this is a byproduct of financial reporting on income, not a disclosure of the underlying tonnage at risk.

Why nobody audits: an honest answer

It is tempting to read the audit gap as concealment, and the internet obliges with theories in which entire depositories are empty. But the more banal explanation fits the evidence better: full physical audits are genuinely expensive and operationally miserable, the institutional incentive to perform one is near zero, and the incentive to avoid one is quietly strong.

Consider what a real audit of a large holding requires. Tens of thousands of bars must be individually weighed — to a fraction of a gram, against a bar list — and a statistically meaningful sample must be drilled or scanned for assay. The work happens in a working vault under security protocols, which means slow. The Netherlands' audit of its gold, one of the more thorough recent examples, took weeks of specialist labour for a 612-tonne holding. Scale that to the US position and you are contemplating a multi-year, multi-million-dollar programme.

And what does the institution gain? If the audit confirms the books — the overwhelmingly likely outcome — it has spent years and millions to learn what it already believed, and it has simultaneously validated every future demand for repetition. If it finds a discrepancy, it has detonated a confidence crisis in its own currency arrangements. Heads, nothing gained; tails, catastrophe. Under that payoff structure, the rational bureaucratic choice is seal verification, custodian attestations and careful language. The gap persists not because anyone is hiding an empty vault, but because no one with the power to order an audit has an incentive to.

• Cost and time: weighing and assaying tens of thousands of bars is a multi-year industrial exercise for a large holder.
• Precedent risk: one clean audit invites demands for audits on a schedule, forever.
• Downside asymmetry: a discrepancy found is a crisis authored by the auditor; a clean result merely confirms the status quo.
• Operational exposure: opening compartments and moving bars increases, however slightly, the physical risk the procedures exist to eliminate.

What a credible modern audit would look like

Strip away the politics and the exercise is not exotic. Private bullion vaults, mining companies and ETF custodians already run programmes that combine the elements a sovereign holder would need, and none of the individual components requires new technology, new treaties or new law — only a decision that the exercise is worth doing.

A public, current bar list

The single cheapest and most powerful step is the one Germany has already taken: publish the full inventory — serial number, refiner, gross and fine weight, assay, vault location — and update it as holdings change. A public list does not itself prove the bars exist, but it converts an unfalsifiable claim into a falsifiable one: outside parties, refiners and researchers can cross-check serial numbers against known production records, and any inconsistency becomes visible rather than buried in an internal ledger.

A rolling, independently witnessed sample

Rather than a single traumatic full recount, a holder could commit to auditing a fixed percentage of its holding — five or ten per cent — every year, on a schedule that cycles through the entire inventory over a decade, with an outside accounting firm or an international body such as the Bank for International Settlements invited to witness rather than merely receive a report. This spreads the cost, avoids the all-or-nothing political risk of a single dramatic count, and — crucially — keeps the verification current instead of freezing it at a single historic date.

A separate line for encumbered gold

Reserve statistics could distinguish, as ordinary corporate accounts must, between gold held free and clear and gold currently out on lease or pledged in a swap. This requires no new counting — central banks already know their own lending book — only a decision to publish a number that today is treated as commercially sensitive.

Taken together, these three steps would not require melting a single additional bar beyond what a rolling sample calls for, and they would close most of the gap between 'reported' and 'verified' within a single budget cycle. That no major holder beyond Germany has adopted any of them, more than a decade after the Bundesbank's programme demonstrated both the method and the absence of any adverse finding, is itself a data point about how the institutions involved weigh the costs and benefits of being believed.

What would settle it

The unsatisfying truth is that certainty in this market is achieved the German way: slowly, bar by bar, with a furnace. But between full recasting and the current state of affairs there is a wide middle ground that any holder could occupy tomorrow, at modest cost, and the fact that none do is itself information.

A holder could publish a complete, current bar list — serial numbers, refiners, weights, fineness, vault location — as the Bundesbank now does, which allows any inconsistency to be spotted publicly. It could invite an external auditor to witness a rolling physical verification, a few percent of bars per year, so the entire holding is re-proven on a decade-long cycle rather than frozen at 1953. It could report leased and swapped tonnage as a separate line, as accounting standards would demand of any commercial balance sheet. And custodians could permit owners' inspectors into the vault on a routine basis, which a few now do and most still do not. None of this requires new technology or treaties. It requires a decision that verification is a public good rather than an institutional risk.

Until that decision is taken somewhere major, the 36,000-tonne figure will remain what it is today: almost certainly true, and not quite proven. For the gold market this is an oddly fitting foundation. Gold's entire monetary appeal rests on the idea that it is the asset that does not depend on anyone's promise — and the world's accounting of it rests, at the final step, on promises. The metal in the vault is the one part of the financial system that is supposed to need no trust. The number that describes it requires a great deal.

What readers should conclude

For a private holder, the audit gap is mostly a lesson in epistemics rather than an actionable risk. The scenarios in which official gold is substantially missing are also the scenarios in which the revelation is the least of anyone's problems. But three practical readings follow. First, treat all official reserve statistics as reported, not verified, and be appropriately sceptical of any analysis that treats the second decimal place as meaningful. Second, the distinction that matters — allocated, identified bars versus claims on gold — applies with even more force to private investors than to central banks: if a government cannot be certain of metal it cannot see, neither can you. Third, watch the behaviour, not the statements. Repatriation requests, published bar lists, central banks that melt and re-assay — these are the actions of institutions converting trust into knowledge. When the owners of gold behave as if verification matters, they are telling you that it does. Fourth, resist the temptation to convert an audit gap into a conspiracy: the absence of proof is not evidence of absence, and the balance of institutional behaviour — cautious lending, careful custodial procedures, a repatriation programme that found nothing wrong — points toward an official sector that is, in the main, exactly as solvent in gold as it claims to be. The honest position is neither blind trust nor reflexive suspicion, but a calibrated scepticism: believe the number provisionally, understand precisely what kind of evidence it does and does not rest on, and treat any holder that voluntarily narrows the audit gap — through published bar lists, rolling verification or melted-and-recast repatriation — as more credible than one that simply asks to be believed.