The Unsecured Physical Layer: A Canadian Kidnapping Case and Crypto's $5 Wrench Problem

The inventory is what stays with me. Two thousand rounds of ammunition. A suppressor. Firearms that, according to the reporting out of Canada, had no legitimate reason to be inside a private residence — let alone in the hands of three people who had sworn an oath to the state that issued them. Three members of the Canadian Armed Forces have been charged in connection with the kidnapping of a cryptocurrency exchange owner. That is the sentence the wire services ran. It is accurate. It is also almost perfectly useless, because it compresses the two facts that actually matter into a frame we have all been trained to skim past: crypto, again, with the crime. So let me separate them. The first fact is that a person who controls a large pool of cryptographic assets was taken. Not hacked. Not drained by a compromised bridge, not rugged by a governance vote at 3 a.m. Taken. The oldest attack in the ledger, executed against the newest asset class on earth. The second fact is where the bullets came from. Almost all of the seized ammunition was, according to the reporting, allegedly stolen from the Canadian military itself. That second fact should have led every security newsletter on the planet, and it is the one that evaporated fastest from the cycle. I understand why. 'Crypto exchange owner kidnapped' fits a template already sitting on the shelf. 'A G7 military's ammunition supply leaked into a criminal enterprise' fits no template at all, and templates are what newsrooms run on at two in the morning. I don't write for the template. Finding the signal in the static here means ignoring the headline and reading the inventory, because the inventory tells you what kind of attacker showed up. A man with a knife is opportunistic. A man with a suppressor and two thousand rounds of state-issued ammunition is something else entirely. A two-trillion-dollar asset class was attacked this quarter. The attack surface was a human wrist. Nine years I have spent writing about cryptographic guarantees — threshold signing, air-gapped key material, hardware roots of trust — and I have never once read a threat model from a major protocol that terminated in a person's body. That is the gap. Not a bug in the code. A hole in the imagination. To understand why this case is a data point and not an outlier, you have to go back to a webcomic. In 2009, Randall Munroe published strip 538. Two characters discuss how to break into a laptop. The punchline is not the encryption; it is the wrench. Why spend a decade on cryptanalysis when you can simply find the person who knows the password and hit them until they tell you? The joke has aged into doctrine. The industry now calls it the $5 wrench attack, and I have come to think the price is the point — the number is a taunt aimed at every security budget in this business. Run the arithmetic the way an attacker runs it. Breaking secp256k1 is not a research problem, it is a physics problem, and physics is not currently accepting bids. But a rented van, a disposable phone, three men willing to hurt someone, and a week of surveillance on a public figure? That is a line item. It fits on a napkin. The asymmetry is not a flaw in the cryptography; it is the cryptography working exactly as designed and thereby redirecting the attacker to the cheapest available path. The most useful artifact in this entire field is not a whitepaper. It is a spreadsheet. Jameson Lopp has maintained a public ledger of known physical attacks against cryptocurrency holders since 2014, and by his running count there have been several hundred confirmed incidents. The curve is not flat. It bends upward through 2021, bends again through 2024, and has not come back down. Read the entries in order and a pattern emerges that no consensus mechanism can address. The early cases were home invasions against people who had talked too loudly in forums. Then the attacks professionalized. Attackers began arriving with information — names, addresses, portfolio sizes, the school run, the gym schedule. By 2025 the pattern had mutated again: a wave of abductions in France targeted not the founders themselves but their parents and their children. That is the tell. When the family becomes the target, the attacker has already solved the technical problem and moved on to the human one. Canada had its own cluster of publicly reported incidents in 2024, in Ontario and British Columbia, that now reads like a rehearsal for the charges described this month. What does an exchange owner actually control? This matters, because the answer determines what a kidnapper thinks they are buying. In a mid-sized exchange, the founder typically holds one share of a 2-of-3 or 3-of-5 signing quorum, with the other shares in geographically separated safe deposit boxes, or held by co-founders, or parked with a qualified custodian. They may also hold administrative authority over the internal ledger — and in most centralised venues, the internal ledger is the chain as far as the user is concerned. Add the authority to authorise large OTC settlements and you have a single human being who is, functionally, a hot wallet with a mortgage. This is sold to users as institutional-grade multisig. Technically, it is. It is also a map. Three humans. Three cities. One of whom has a headshot on the company's about page. The legal architecture around the case is worth understanding too, because it explains the silence. Kidnapping in Canada falls under section 279 of the Criminal Code and carries a maximum of life imprisonment. A suppressor is a prohibited device under section 84(1), with possession and trafficking offences under sections 95 and 96. Where serving members are accused, there is a concurrent-jurisdiction question between the civilian courts and the military justice system under the National Defence Act, and section 130 of that Act can convert a Criminal Code offence into a service offence. Canadian courts can also impose publication bans under section 486.5. That is why you will not read the exchange's name here — not because it cannot be found, but because a court may have ordered otherwise, and because the victim's identity is itself a security artifact. Naming the exchange in a kidnapping case is not journalism. It is target intelligence for the next crew. The attack tree nobody drew Every serious protocol publishes a threat model, and they are genuinely good. Eclipse attacks. Long-range attacks. Validator collusion. MEV extraction. Supply-chain compromise of build tooling. Formal verification of consensus safety properties. There is real intellectual firepower in these documents. There is no section in any of them titled: what happens when someone puts a gun in the founder's mouth. Look at how the security budget is allocated across this industry. Audits run from fifty thousand to half a million dollars and are treated as table stakes. Bug bounties scale into seven figures. Formal verification teams are hired at the seed round. And the physical security line item, for the person who can authorise the transfer of every customer asset on the platform, is frequently a doorbell camera and a dog. The network layer is now genuinely expensive to attack. That is not a failure of the industry — it is the industry's single greatest triumph. And it is precisely why the attack migrated. You do not rob a bank that has replaced its vault with a mathematically unbreakable one. You rob the branch manager. Compliance built the target list Here is the structural insight I most want to plant, and I want to be precise, because it is easy to misread as an argument for opacity. It is not. Regulatory transparency regimes are, incidentally, targeting databases. They were built to prevent money laundering and terrorist financing. They function as a directory of who controls the money. In Canada, the FINTRAC registry of money services businesses is public and searchable. Corporate registries list directors by legal name and often by address. Licensing applications in most jurisdictions disclose beneficial ownership. Layer on top of that the exchange's own about page, the founder's conference bio, the podcast appearance where the guest casually mentions that they personally hold the cold storage quorum, the panel photo where four people stand under a banner reading the name of the venue. I have sat in rooms where a founder was introduced by name, exchange, and assets under management to an audience of four hundred people holding phones. Every one of those phones was a camera. Every one of those cameras was a map. The paradox is exact: the more compliant the jurisdiction, the better the target list. KYC the institution and you have doxxed the human. Nobody has ever run the threat model on the disclosure side of the compliance stack, and until someone does, every new registration requirement quietly improves the intelligence available to people who own suppressors. Why your multisig doesn't save you The reflexive answer to wrench attacks is multisig, and it is a good answer to a different question. Threshold signature schemes — FROST for Schnorr, the GG18 and GG20 families for ECDSA — solve a real and important problem: no single device, no single key, no single custodian can unilaterally move funds. This is genuine progress. I have written about it approvingly and I stand by that. But the coercion is applied to the human, not to the key. A 3-of-5 quorum is not a cryptographic barrier to a person who is willing to visit five houses. It is a list. Social recovery is worse, and it is worse in a way that nobody says out loud. Guardian-based recovery schemes are elegant engineering — they solve the lost-seed problem that has destroyed more wealth than every hack combined. They also convert a key-recovery mechanism into a physical attack graph. Your guardians are now a target list, and they are usually your friends, who did not sign up to be a target list. I watched this exact argument surface during the human-in-the-loop hackathon I ran in 2025, when a participant asked what happens if the recovery quorum is coerced rather than the primary. Nobody in the room had an answer. That silence is the state of the art. Duress features exist. Hardware wallets ship duress PINs. Safe manufacturers sell coercion codes. They fail in practice for reasons that are obvious once you say them: they require the victim to be calm, to remember which code is which, and to possess a decoy balance that survives scrutiny. A duress wallet holding four hundred dollars does not survive contact with an attacker who knows the founder of an exchange and has read the same articles you have. Time locks are the most underrated control in the box. A forty-eight-hour withdrawal delay on large transfers buys response time — genuinely useful, structurally sound. Almost no centralised venue implements it on hot wallets, because users demand instant withdrawals and will migrate to the venue that offers them. The UX of safety is slowness, and this market has never once paid for slowness. What a coercion-aware custody stack actually looks like is not mysterious. Rate-limited panic drains to a pre-authorised cold address. Dead-man switches that alert an external response firm on retainer. Plausible-deniability tiering where the duress credential unlocks a real, funded, smaller account rather than a fake one — attackers test, and a fake account is a death sentence. Geographic separation of humans, not merely of keys. And an actual human security officer on payroll, which almost no exchange below five billion dollars in volume has. Based on the custody reviews I sat through with former audit partners while building the Trust, but Verify series in 2024, I can tell you what those documents said. The quorum was documented. The geography was documented. The incident response plan for a physical threat to a key holder was a single line item that read: contact legal. The insurance gap Two products exist that should meet in the middle, and they do not. Digital asset custody insurance covers theft, hack, and insider misappropriation. It is underwritten by a small pool of Lloyd's syndicates and specialist carriers, and it has matured considerably since 2022. Kidnap and ransom insurance — K&R, a product that has existed for decades for executives, journalists, and high-net-worth families — covers abduction, ransom negotiation, and extraction. The custody policy excludes physical coercion. The K&R policy excludes the transfer of digital assets under duress. There is a seam between the two, and this case sits precisely in it. The reason is actuarial, not conspiratorial. Underwriting requires a loss run. The closest thing this industry has to a loss run is a volunteer-maintained spreadsheet. You cannot underwrite a risk you cannot price, and you cannot price a risk nobody has bothered to count. Insurers are the only people in finance who are paid to be paranoid about exactly the right thing, and when they decline to write a policy they are not being lazy. They are telling you the risk is real and the data is absent. The ammunition Two thousand rounds is not a shoebox. It is a case, or several cases, moved across distance, stored somewhere, and never once missed by the people who were supposed to be counting it. In Canada, a suppressor is not an accessory. It is a statement of intent. Suppressors exist for one purpose, and that purpose is to ensure the sound of what you are doing does not travel. Military ammunition accountability is, in principle, one of the more rigorous chain-of-custody regimes a state operates. Serialised lots. Range reconciliation. Magazine counts signed by two people. Losses happen — they always do — and they are supposed to be reported, investigated, and prosecuted. A loss of this scale is not a clerical error. It is either a systemic accountability failure or an insider network, and the difference matters enormously to everyone downstream. Crypto readers should care about this more than anyone, because supply-chain security is our discipline. We spent a decade learning that the weakest link in a build pipeline is not the compiler, it is the person with commit access. The same logic applies to a magazine: the weakest link is the person with the key to the door. If the reporting holds, a state-grade logistics chain leaked into a criminal enterprise that was targeting our industry. That is a threat-model escalation. It changes the attacker profile from opportunistic to resourced, and resourced attackers replicate. The narrative tax Compare three headlines. 'Cryptocurrency exchange owner kidnapped.' 'Jeweller kidnapped.' 'Hotel owner kidnapped.' In the second and third, the profession is context. In the first, the profession is the story. That asymmetry has a cost, and I think of it as a narrative tax — a premium this industry pays on every crime committed by or against anyone with a crypto label attached. Each incident becomes an argument for more surveillance, more friction, more reporting requirements, and the argument is politically free because the public already holds the prior. Then the new requirements generate new public registries, which generate better target lists, which generate more incidents. That is a loop, not a line, and nobody is standing at the exit. There is a deeper discomfort underneath it. This is the same dynamic that turned Bitcoin from a peer-to-peer cash experiment into a Wall Street instrument. The asset got institutionalised. The humans holding it did not. Here is where I part company with most of the commentary I have read on this case. The consensus response is a supply-side answer to a demand-side problem, and it is shaped like the victim. Hire security. Use multisig. Stop flexing on podcasts. Do not post your car. All of it is reasonable, and all of it asks the target to absorb the cost of a structural failure. We do not tell bank tellers to buy their own armoured vests. The uncomfortable thing the ETF era actually did — and I say this as someone who spent 2024 writing about custody because the spot Bitcoin ETF made custody the industry's central question — is that it solved physical security for institutions and abandoned it for individuals. BlackRock's custodian has vaults, guards, insurance, and a legal department that will sue you into the ground. The founder of a mid-tier Canadian exchange has a 2-of-3 and a Ring doorbell. We built a two-tier security system and called the top tier adoption. Nor does the compliance layer rescue anyone. A compliance-first stablecoin issuer can freeze an address within a business day, and that is a real capability. It is aimed at the wrong moment. Freezing happens after the asset moves. Kidnapping happens before. The compliance layer protects the issuer's balance sheet, not the human being, and it cannot protect the human being, because the ransom demand is increasingly not denominated in a token anyone can blacklist. It is denominated in cash, or gold, or a flight. And this industry — which will spend eight figures subsidising total value locked with token emissions, which can measure liquidity to the basis point, which builds dashboards for everything — has never once budgeted a line item for the safety of the people who hold the keys. That is not a resource problem. It is a priorities problem wearing a resource problem's clothes. Three things I will be watching, in order of how much they would tell us. Whether the charges proceed under the Criminal Code or through the National Defence Act. A civilian prosecution says the state treats this as a kidnapping. A service offence says the state treats this as an internal supply-chain failure that happened to escape the wire — a very different institutional admission, and the one that would actually trigger an ammunition audit. Whether the Canadian Forces announce a reconciliation of their stocks. If two thousand rounds can leave without a flag being raised, the number that left undetected is the number that matters. And whether a K&R product gets written for crypto founders within the next twelve months. I have heard three separate conversations about it since the France wave and not one term sheet. The day an underwriter prices this risk is the day this industry admits it has a physical layer. The next narrative in this business will not be about tokens. It will be about the custody of humans — the layer every whitepaper skipped, every audit excluded, and every conference badge quietly published. Finding the signal in the static of this particular wave means accepting something genuinely uncomfortable: the cryptography worked. The humans didn't. And no upgrade path fixes that except the one nobody wants to fund, because it does not produce a yield. So ask yourself, the next time you read a founder's name and their exchange in a conference programme — who is protecting that person, and what is the plan for the forty-eight hours after someone decides the answer is nothing?