The word insured does a lot of quiet work in a custody pitch. A custodian will tell you they carry some large policy, the number gets screenshotted, and everyone moves on feeling covered. Then something goes wrong that the number never had anything to do with, and people are genuinely shocked to learn their coins were never insured against the thing that actually happened to them. I have gone through enough of these disclosures to have a rough sense of where the gap between the headline and the reality usually sits, and it is almost always in the same three or four places.
So it is worth slowing down on what these policies really are. Custody insurance is not one product. It is usually a stack of separate coverages, each with its own trigger, its own limit, and its own long list of things it politely declines to pay for. If you only remember one thing, remember that the coverage is written around how the custodian holds the assets, not around how you might lose them.
Crime and specie, and why the distinction matters
The two policy types you will see most are crime coverage and specie coverage, and they answer different questions. A crime policy is built around dishonest acts. Employee theft, fraud, an insider walking off with keys, that kind of thing. Specie coverage comes out of the world of insuring physical valuables in a vault, gold bars and fine art and cash, and it gets applied to hardware that sits in cold storage. Think of specie as insuring the box and crime as insuring against the people with access to the box.
What that combination handles well is a fairly narrow scenario. Keys held offline, in a physical facility, stolen or destroyed by someone on the inside or a break-in from the outside. That is a real risk and it is good that it is covered. It is just a small slice of the ways money actually leaves people's accounts. Most losses I have watched happen had nothing to do with a vault being cracked.
The exclusions are where the education happens. Read them and you will typically find carve-outs for loss of private keys where no theft occurred, for insider collusion above a certain threshold, for war and nuclear events which sounds silly until a jurisdiction gets messy, and very often for anything touching hot wallets, which is to say the wallets the exchange actually uses to move your funds around. The assets sitting in cold storage are the well-insured ones. The assets in motion, the ones being used, are frequently the least covered, and those are the ones involved in most operational failures.
The losses the policy is not looking at
Here is the part that catches people. A custody policy is designed to protect the custodian's holdings from theft. It is generally not designed to make you whole for these:
- Exchange or custodian insolvency. If the business fails and there were never enough assets to go around, an insurance policy against theft does not fill that hole. This is the big one, and it is exactly the scenario most people assume they are covered for.
- Smart contract failure. If funds move through a protocol and a bug or exploit drains a contract, that is usually not a theft from the custodian's vault, and specie coverage has nothing to say about it.
- Your own account being phished or SIM-swapped. If an attacker convinces the platform they are you, or convinces you to hand over a code, many policies treat that as authorized activity. The keys were never stolen from the custodian. You gave access away.
- Depegs, protocol de-risking, and market loss. Obvious, but worth stating. Insurance covers theft and physical loss, not the price going down or a stablecoin losing its peg.
Notice the pattern. The policy protects the custodian's operational integrity. It does very little for the failure modes that live at the edges, in the business itself, in the code, and in your own login. Those edges are where I have seen the actual money disappear.
Per customer versus aggregate, the sentence to hunt for
Even when a loss is covered, the limit is rarely what it looks like. A custodian advertising a large policy is almost always quoting the aggregate limit, meaning the total the insurer will pay across the entire pool of customers for a single event. If a facility gets hit and thousands of clients are affected at once, that number gets divided, and it can get divided down to something that no longer feels like coverage at all.
The number you actually care about is the per customer sub-limit, if one exists, and often it does not. When it does, it can be surprisingly modest relative to what a serious holder keeps on the platform. So the mental exercise is to stop reading the headline as your coverage and start asking what portion of the aggregate would realistically reach your account in a bad event. Frequently the honest answer is a small fraction, and sometimes there is no per customer floor at all, which means you are sharing one pool with everyone else and hoping the event is small.
How to actually read a custodian's disclosure
Here is the workflow I use when someone asks me whether a platform is safe to park size on. It takes maybe twenty minutes and it filters out most of the wishful thinking.
- Find the actual policy language, not the marketing page. If the only place the insurance appears is a homepage banner and there is no document you can read, treat the coverage as roughly zero until proven otherwise.
- Identify which coverage types are in the stack. Crime, specie, both, something else. Then check what share of assets is held in cold storage versus hot wallets, because the insurance mostly follows the cold storage.
- Read the exclusions before the coverages. It is faster to learn what is not covered, and the exclusions tell you the real shape of the policy.
- Find the word aggregate and the word per and see which one attaches to the big number. If the big number is aggregate and there is no per customer sub-limit, mentally shrink it.
- Ask who the named insured is. If the policy names the custodian and you are not a beneficiary, a payout goes to the company, and you are a creditor waiting in line behind everyone else.
That last point is the one people miss most. Being covered by a policy your custodian owns is not the same as being paid by it. In an insolvency the insurance proceeds can become an asset of the estate, and you are back to standing in the creditor queue, which is precisely the situation you were trying to insure against.
What to do with all this
None of this means custody insurance is worthless. Insider theft from cold storage is a genuine risk and it is good to have it covered. It just means the word insured is answering a much narrower question than the one you are probably asking. The practical move is to size your exposure to any single custodian based on the failure modes the policy does not touch, insolvency and phishing and smart contract loss, because those are the ones that will actually be your problem.
For balances you cannot afford to lose, spreading across venues and keeping the largest holdings in self-custody does more real work than any headline policy figure. And when I am comparing platforms on Blockcircle or anywhere else, I read the custody disclosure the same way I read a counterparty's balance sheet, as a document written to make a case, with the interesting information sitting in the parts they did not put in bold. Assume you are still exposed to whatever the exclusions list, and you will be right more often than the number on the banner suggests.