1,250% is not a typo. It is not a fat-fingered decimal in a leaked spreadsheet. In the Bank of Russia's draft capital framework for crypto exposure, a bank's own positions โ and any client position where the bank carries the loss โ attract a risk weight of one thousand two hundred fifty percent. Push that through the standard 8% minimum capital ratio and the arithmetic resolves to a capital requirement equal to 100% of the exposure. One ruble of capital for every ruble of crypto. No leverage. No balance-sheet room. No path to a spread.
In the same document, client positions the bank does not stand behind carry a 50% risk weight. That is a 4% capital charge. Two coefficients, twenty-five times apart in economic effect, printed in the same table. That asymmetry is the policy. The "1% cap" that dominates the headlines is downstream of it โ a constraint on the size of a business that the weight table has already made commercially irrational.
Let's look at the data before we look at the narrative.
Logic prevails where hype fails to compute.
I've spent years reading capital frameworks the same way I read unverified Solidity โ hunting for the line that decides everything. In 2017 I spent sixty hours inside the source of "Ethereum Gold," a fork promising enhanced throughput, and found an integer overflow in the mint function that allowed infinite supply under specific block heights. The marketing was clean. The code was not. The lesson I took from that episode is the one I apply here: never evaluate a system by its stated intent. Evaluate it by the arithmetic it actually enforces.
Where This Rule Actually Lives
The proposal sits in the Bank of Russia's prudential rulebook, not in a crypto law. That distinction matters more than it sounds. This is a bank-capital instrument โ the same family of rules that governs credit risk, market risk, and sovereign exposure. It is scheduled for publication around the fourth quarter of 2026, taking effect ten days after publication, with institutions reporting turnover and the two new ratios beginning January 2027.
The architecture is two ratios. N31 measures crypto exposure against a single credit institution's own funds. N32 measures it against group capital on a consolidated basis. The design intent is visible in the denominators: the entity that bears the risk is the entity whose capital must absorb it. A group cannot inflate its tolerance by parking exposure in a lightly capitalized subsidiary, because the consolidated ratio re-captures it. Anyone who has watched a bank shuffle risk between a trading entity and a funding entity will recognize this as a deliberate anti-structuring clause. The drafters have been in this fight before.
The numerator is where the engineering gets serious. Covered exposure is not limited to spot holdings. It reaches direct and indirect investments, derivatives whose price is linked to crypto, and loans, bonds, guarantees, repos, and credit lines whose settlement or value depends on crypto. Foreign digital instruments are swept in as well.
Read that list again, slowly. The draftsman did not simply count coins on the balance sheet. The draftsman enumerated every mechanism by which a bank could acquire crypto-linked economics without holding the asset, and then closed each one. A total return swap that converts crypto price exposure into a fee stream is captured. A guarantee written against a client's token holdings is captured. A repo facility collateralized by digital assets is captured. In 2017, the projects that died did so because their code allowed an assumption nobody had tested to propagate. The Russian draft is the opposite posture: it assumes evasion and writes the rule against it.
None of this is original, and it does not pretend to be. The 1% cap on Tier 1 capital and the 1,250% weight map almost exactly onto the Basel Committee's 2022 treatment of what it calls Group 2 crypto assets โ the bucket that excludes tokenized traditional assets and captures everything from Bitcoin to the long tail. Basel set a 1% Tier 1 exposure limit for Group 2 and assigned a 1,250% risk weight to the sub-bucket considered least able to demonstrate effective hedging. Russia's draft is a high-fidelity localization of that standard, with two local additions: an explicit custody carve-out and the inclusion of foreign digital tools.
That is the skeleton. Now the load-bearing beams.
Reading the Coefficients
Start with the 1,250%. A risk weight is a multiplier. The capital a bank must hold against an exposure equals the exposure multiplied by the risk weight, multiplied again by the minimum capital ratio. At 8%, a 1,250% weight produces 1.0 โ full deduction. The exposure is removed from capital, in economic substance, as completely as if the bank had written the position down to zero on the day it was booked.
This is why the "harsh 1% cap" framing misses the mechanism. The cap limits scale. The weight limits existence. A mid-sized Russian bank with 200 billion rubles of own funds can hold two billion rubles of proprietary crypto before breaching N31. At prevailing prices, that is a rounding error on a trading book, and it must be funded with capital that earns nothing and sits against an asset that the risk function cannot hedge. The rational choice is not to hold a small position. The rational choice is to hold none.
The 1,250% weight is not a constraint on proprietary crypto trading. It is a prohibition that has been routed through a coefficient so that nobody has to write the word "ban."
Now flip to the 50%. Client positions where the bank carries no loss obligation get half the weight, which is a 4% capital charge. A bank custodying 100 billion rubles of client crypto with no retained liability needs four billion rubles of capital against it. That is real, but survivable. The bank books a fee, not a position. It becomes a pipe, not a principal.
The pivot between those two outcomes is a single legal question: who bears the loss if the asset is seized, frozen, or otherwise impaired? The draft treats that determination as the boundary between "the client's problem" and "the bank's problem." Get classified as non-liable and the position costs 4% in capital. Get classified as liable and it costs 100%.
I want to be precise about how fragile that pivot is. The classification depends on an off-chain legal characterization โ a fact about contracts, custody terms, and jurisdiction, not a fact any chain can report. In the systems I audit, we call something like this an oracle problem. The rule's critical variable is an assertion about the world that must be adjudicated by a human institution, slowly, with disputes. Legal oracles are not deterministic. They are contested, revisable, and, under stress, exactly as reliable as the least aligned party in the dispute.
And stress is the whole point. A capital rule is a claim about what happens in a bad state of the world โ when prices gap, when counterparties fail, when a jurisdiction freezes assets. Consider the Russian context specifically. Client crypto held in custody at a Russian bank is, for a meaningful share of users, a vehicle for cross-border settlement and store of value under sanctions. That is precisely the kind of asset that gets frozen. The moment a freeze becomes a legal contest, the question of who bears the loss stops being theoretical, and so does the gap between a 4% and a 100% capital cost. A custody book that looked cheap in calm conditions can reclassify wholesale in a crisis. That is not a hypothetical failure mode. That is the failure mode the rule is nominally built to survive, relocated into a classification decision.
The hedging rules carry the same conservative DNA. Netting is permitted only within qualified lower-risk categories, and only subject to conditions covering the asset, settlement, maturity, and freezing and liquidity risk. Direct holdings and higher-risk exposures cannot be fully offset by hedges.
Compare that to what other jurisdictions are doing. Canada's 2027 capital rules provide relief for positions hedged across exchanges โ an acknowledgment that crypto liquidity is fragmented and that a hedge executed on venue A is still a hedge against the position on venue B. Russia's draft declines that acknowledgment. It requires the offsetting legs to sit inside the same qualified bucket under the same conditions. In practical terms, that means the only netting that survives is the netting that was never very useful.
There is a second-order effect here I want to name, because I have watched it happen in other systems. When a rule conditions risk relief on settlement, maturity, and freezing characteristics rather than on price correlation, it is effectively telling banks that hedging is a settlement and custody problem dressed up as a market-risk problem. I spent three months during DeFi Summer modeling the gap between what Compound and Aave considered a valid price and what Uniswap and Sushiswap were actually quoting. The lesson was not "oracles are hard." The lesson was that a hedging exemption is only as good as the latency budget you can enforce around it. Russia's draft sets the latency budget close to zero by refusing cross-venue relief. Under that constraint, the bank's cheapest response is not to hedge. It is to not hold.
That is the design talking. And the design says: custody yes, principal no.
Run the business case end to end. A bank wants to offer crypto exposure to its clients. Two structures are available. Structure one: the bank buys and holds the asset, or writes a structured deposit linked to its price, or runs a delta book against client flow. Capital cost: 100% of the exposure. Structure two: the bank holds the asset for the client, assumes no loss liability, and charges a custody fee. Capital cost: 4%. The spread between the two is twenty-five times. No treasury committee looks at that and picks structure one. The rule does not need to forbid proprietary crypto activity. It prices it out of existence and lets the market deliver the verdict.
One more coefficient deserves attention, because it is the one that looks boring and is not. The 50% custody weight is higher than the weight on cash and sovereign debt, which is zero. That tells you the regulator is not treating crypto custody as a neutral, client-driven service. It is treating it as a medium-risk activity that requires a cushion. Set against the 1,250% proprietary weight, 50% looks generous. Set against zero, it is a signal. Custody is permitted, but it is not free, and the drafters want banks to remember that every ruble of client crypto adds a permanent claim on capital that no other custody asset does.
Which raises the fee question. Custody is a thin-margin business even without a capital drag. Add a 4% capital charge per unit of client assets under custody, and the fee schedule must clear a hurdle that digital-asset custody desks in more permissive jurisdictions do not face. If the compliant channel cannot be priced competitively against offshore alternatives, the capital relief in the custody bucket is theoretical. The rule would then have created a permission nobody uses. That is a common outcome in prudential design, and it is rarely reported, because a permission unused is not a headline the way a permission granted is.
Then there is the numerator's long arm, which reaches indirect investments and foreign digital instruments. Read that literally. A bank holding exposure to an offshore stablecoin, or to a digital instrument issued outside Russia, is captured. This is where a prudential rule acquires a capital-control accent. The framework is not only measuring how much crypto risk sits on a bank's balance sheet; it is measuring how much of that balance sheet is connected to crypto rails Russia does not control. Under sanctions, that is not a neutral measurement. It is a mapping exercise, and a map is the first thing you draw when you intend to govern territory.
I have to be careful about confidence levels here, because this is where analysis shades into inference. The mapping to Basel's Group 2 parameters is solid โ the numbers line up too cleanly to be coincidence. The capital-control reading of the foreign-instrument clause is a reasonable inference, not documented intent. I flag it as such. The pattern โ a rule that is formally about solvency but functionally also about reach โ is one I have seen before, and it usually means the regulator is optimizing for two objectives and disclosing one.
Logic prevails where hype fails to compute.
The Protection Is Real, and Conditional
The headline defense of this framework is that a harsh 1% cap protects bank customer assets. The logic chain is clean: confine crypto risk to a sliver of bank capital, and bank capital survives any crypto outcome; a bank that survives does not need to be bailed out; a bank that is not bailed out does not have its deposit base impaired. Protection, in this reading, is a property of the bank, and it reaches the customer indirectly.
That chain holds. I am not going to pretend it doesn't. But it holds with a precondition the headline erases, and the precondition is the same legal oracle described above. Client assets are protected under this framework only when two things are simultaneously true: the bank genuinely does not bear the loss, and the institution that does bear the loss is solvent and competent. Break either one and the 50% weight stops being a protection and starts being a discount.
Here is where my own audit history becomes relevant. After the 2022 collapse I spent six months inside the emergency governance contracts of Terra Classic โ the failsafe machinery that executed the hard fork. I found that the pause function depended on a single multisig. One wallet. The decentralization claims were, in that specific and load-bearing place, decoration. The failure mode was not malicious. It was structural. Every fail-safe that routes through one custodian has converted a distributed problem into a single point of failure, and the more central the fail-safe, the more concentrated the exposure.
Now apply that lens to the second half of the Russian arrangement. The draft contemplates digital depository institutions and, alongside the rulemaking, an arrangement in which a state-linked bank holds the keys. Read charitably, that is public infrastructure for custody. Read structurally, it is a multisig with one signer and a government's name on the wallet. Every custody asset in the system would then depend on one institution's key management, one institution's internal controls, one institution's interpretation of the loss-liability question.
The 1% cap protects bank capital from crypto volatility. It does nothing to protect client assets from the failure of the custodian the state has selected.
That is the blind spot. The framework is rigorously decentralized in the dimension that threatens bank balance sheets and rigorously centralized in the dimension that holds client assets. The two designs are not contradictory โ they serve different objectives โ but the protection narrative only covers the first one, and the coverage is treated as complete.
There is an operational consequence I want to sketch, because it comes from the part of my work that sits at the intersection of machine learning and on-chain systems. Both N31 and N32 depend on classification: which exposure is proprietary, which is client, which is non-liable, which is foreign. Large institutions automate classification. If that automation runs through LLM-based tooling โ and increasingly it does โ then the classification pipeline becomes an attack surface with a new class of failure modes. Adversarial prompt injection against a reporting assistant is not a memory-safety bug. It is a mislabeling of loss liability that moves a position from the 50% bucket to the 1,250% bucket, or the reverse, depending on which direction serves the operator.
I built a sandbox framework for exactly this class of problem last year, where language models generated and tested transaction payloads without risking real funds. The finding that stuck with me was uncomfortable: models that are reliable at classifying clean data are reliably wrong at classifying adversarially framed edge cases, and legal edge cases are precisely the ones that decide capital outcomes. A rule whose critical input is a legal judgment, processed through automated tooling, with a twenty-five-fold capital consequence attached, has exported its hardest problem to the least auditable part of the stack. Reporting begins in January 2027. That is enough time to build the pipeline and not quite enough time to audit it.
The final contrarian point is about where the activity goes. A framework that penalizes proprietary holdings at 100% and declines most hedging relief does not eliminate crypto demand in Russia. Demand is driven by capital controls and cross-border settlement needs, and it does not read capital rules. What the framework does is make the compliant channel expensive and narrow. The plausible equilibrium is a formal banking system that is quiet and a grey, offshore, and peer-to-peer layer that is not. If the compliant channel is genuinely cheaper by 2027, activity migrates inward. If it is not, the rule has succeeded at protecting bank capital and failed at capturing the flow โ and the protection narrative will be true and irrelevant at the same time.
What to Watch, and Why It Travels
It is worth stating plainly what this rule is not. It is not a Russian embrace of crypto, and it is not a Russian ban. It is a template, and templates are the part of regulatory news that outlives the news cycle. Sanctioned and capital-controlled economies face a shared problem: real, persistent domestic demand for assets that sit outside the banking system, and a state interest in seeing that demand without absorbing its volatility. This draft is one answer โ permit the pipe, prohibit the principal, centralize the keys, map the foreign rails. Other jurisdictions with similar constraints will read it. Some will copy it.
That is why the framework matters more than its market footprint suggests. As a regional rule it moves nothing global. BTC does not reprice on a Bank of Russia consultation. But as an artifact of how a major state proposes to institutionalize crypto under duress, it is one of the more instructive documents of the cycle โ and the cycle is a bear market, where the useful question is not what goes up next but which plumbing holds when it is squeezed. This plumbing is designed to hold bank capital and to hold client flows. It does not claim to hold client assets through a custodian failure, and the reporting will not measure that either.
The number to watch is not 1%. It is 1,250, and what the final text does with the liability boundary. The draft publishes around the fourth quarter of 2026 and takes effect ten days later, with N31 and N32 disclosure from January 2027 โ a window long enough for the coefficients to move and for the key-custody arrangements to be specified or quietly dropped. Everything about the practical outcome hinges on that specification. If the state-linked custodian takes the keys and the liability classification stays legal rather than mechanical, Russia will have built a system that insulates bank capital with genuine rigor and concentrates client risk with equal rigor, and both facts will be reported as one story about prudence.
Which one is the protection โ the capital cushion or the custodian? The framework answers that question in a coefficient most coverage will never print. Logic prevails where hype fails to compute.