Agio Ratings

For an entity that files no accounts, there is no other way to produce a defensible default forecast. Calibrated statistical models combine on chain signals such as reserve balances, wallet flows and transaction patterns with conventional financial indicators to produce a numerical probability of default, refreshed daily across more than 70 counterparties. A further layer converts a portfolio of those probabilities into a loss distribution. Take the models away and what remains is reputation and regulatory status, which the company identifies as exactly what risk managers were relying on before.

The product is explicitly an input rather than a decision. Ratings are described as quantitative inputs for exposure limits, counterparty selection and insurance underwriting, fitting into the same framework governing traditional credit decisions, and the named customer describes them as an independent signal alongside its own monitoring rather than a replacement for it. That framing is correct for a ratings business, where the user sets policy and the rating informs it. No threshold or automated action mechanism is described, which is consistent with the model.

The validation evidence is out of sample, real world and tested in both directions. The models flagged a high probability of default at FTX four months before its bankruptcy, a correct positive on the most consequential failure the market has seen. They separately judged that Bybit had the resilience to survive its $1.5 billion security breach, a correct negative under extreme stress.

Rather than accuracy on its own benchmark, the company has two public calls that could have gone against it. The methodology is published, with ratings driven by nearly twenty scored variables and the categories named. Forecasts are described as empirical and calibrated, and specific probabilities are published with quarterly change.

Users are named: Wintermute, Ledn and LMAX Digital among trading firms and lenders, with a trading firm's risk lead quoted describing the alerts as an independent signal alongside internal monitoring. The strongest commercial validation is Relm Insurance's FALTAWEB3, an exchange default product built on Agio's ratings, which means an insurer is pricing real risk on these numbers rather than consuming them as research.

More than $11 million has been raised, most recently a $6 million round in October 2025 led by AlbionVC with Portage Ventures and MS&AD, the insurance group's venture arm, participating. Coverage extends across more than 70 exchanges and custodians, with banks, insurers and regulators stated among customers.

Clients upload counterparty portfolios to obtain loss distributions, which discloses their venue exposures to the vendor, and those clients compete with one another. No boundary statement is published. Nothing states whether aggregate client exposure data informs ratings or research, how portfolio inputs are segregated, or what the company may infer and publish from seeing where the market's capital actually sits.

Privacy exposure is low, because the subjects of analysis are institutions rather than individuals and the inputs are public chain data and financial indicators rather than personal information. No data protection agreement, retention schedule or subprocessor list is published. What remains undisclosed is how client portfolio and exposure data is handled, which is commercially sensitive since it reveals where a trading firm holds capital.

Agio publishes no attestation, certification, trust center or enumerated framework. Banks, insurers and regulators are stated customers, and their supplier assessments would have covered this ground. For a provider holding client counterparty exposure data, a published control set is the natural disclosure, particularly as a breach would reveal where major trading firms hold capital.

Three regulatory instruments are named as the drivers creating institutional demand, covering the European digital asset regime, United States stablecoin legislation and an accounting bulletin governing crypto custody, and the company argues correctly that banks and regulated funds entering these markets need documented risk frameworks. Regulators are stated to be among its customers.

The material gap is its own status: this is a ratings business and no recognition or registration as a credit rating agency under any regime is claimed, which matters because regulated firms can only use ratings from registered agencies for certain capital and reporting purposes.

No individual is assessed, and the adapted exposure is the classic hazard of the ratings business rather than demographic bias. A published default probability can help cause the outcome it forecasts, because institutions withdrawing capital from a downgraded venue reduce its liquidity and reserves, which the model then observes as further deterioration.

The company publishes named counterparties with numerical probabilities openly, which makes that reflexivity more immediate than for a subscription only rating. Nothing published addresses the feedback effect, how a rated entity is notified before publication, or how a smaller venue with thinner on-chain history is treated relative to an established one.

The affected party is the rated institution rather than a consumer, and no guarantee, indemnity or correction process is published. A published default probability directly affects an exchange's ability to attract institutional capital and to obtain insurance. Nothing describes whether a venue is notified before publication, whether it may submit information or challenge inputs, or how a rating derived from misread on chain reserves would be corrected. Established rating agencies operate formal appeal and comment procedures precisely because the consequences are material.

Input categories are enumerated specifically rather than gestured at, covering on-chain reserves, operating leverage, flow volatility and venue maturity among nearly twenty scored variables, and the company describes combining chain level signals such as wallet flows, reserve balances and transaction patterns with traditional financial indicators. That lets a buyer understand what a rating rests on. No chain analytics provider, market data source or financial data vendor is named, and no subprocessor list appears, so dependency concentration cannot be assessed.

Delivery is as ratings, alerts and portfolio modeling rather than through embedded integration, and the value proposition is explicitly that a firm avoids building models and maintaining data pipelines itself. No named risk system, treasury platform, order management system or interface documentation is published, so how ratings reach a risk team's existing exposure monitoring is not described.

Where data is hosted and processed is not published, with no provider, region selection, residency commitment or private deployment option named. Client portfolio exposures are the sensitive holding rather than personal data. Institutions in Europe and the United States subject to their own operational resilience requirements would ordinarily ask where a risk data provider processes and stores that information.

The ratings themselves are open, with named counterparties published alongside their default probabilities and quarterly movement. Commercial terms are not. There is no published pricing, packaging or basis of charge, and nothing indicates what access to the full daily updating platform costs or whether ratings, alerts and portfolio modeling are sold separately.

The buyer set is complete for a young company, spanning trading firms and market makers, funds, banks, insurers and regulators, the parties on every side of a counterparty exposure. Rated coverage extends to more than 70 exchanges and custodians, with lenders also in scope. The constraint is subject matter. This is counterparty credit risk in digital assets specifically, deliberately narrow, and the company positions that focus as the reason established agencies leave the gap open.