Calvin Risk

• No certification Calvin Risk itself holds appears on any surface probed on 21 August 2026, so a bank's vendor review starts without the assurance artefact the public record would normally carry.
• The material behind the LLM metrics is described only on Calvin Risk's own pages, and no outside source in the reviewed record confirms its size or composition.
• The reviewed sources disclose no contract and no production scale for any named organization, so a reference call could reach a pilot rather than a deployment.
• CB Insights names Naaia and seven more competitors and describes five selling the same inventory and compliance layer, so the difference a buyer can point to narrows to the loss estimate if one of them adds a value-at-risk score.

The EU AI Act gives Calvin Risk's buyers a duty they have to evidence. The company's feature page states that providers and deployers of high-risk systems must put risk and quality management measures in place, and the platform walks a company through a questionnaire to establish whether a system counts as high risk at all.

Inside the enterprise the problem is inconsistency. Calvin Risk's assessment page says model developers in different business units validate their models in different ways and may not report dimensions such as fairness, explainability, accountability or safety. The product opens with an inventory rather than a test, and the vendor frames full visibility of a company's AI systems as the foundation for risk management and compliance.

Financial services is where Calvin Risk started. A Forbes contributor article says the company focused initially on customers in that sector, where concern about regulation and the cost of a mis-step is particularly high. [s21, s2, s3, s1, s13]

Calvin Platform does three things. It records the AI models and business use cases a company runs. It calculates quality metrics across six dimensions, which the vendor names as performance, robustness, fairness, explainability, accountability and safety. It then reports a value at risk in euros or dollars for each business use case, covering financial, reputational, legal and disruption losses.

The metric counts are published. The LLM module calculates 6 performance metrics, 11 robustness metrics and 2 fairness metrics. Calvin Risk states that it builds those LLM metrics from 150 million tokens of questions, retrievals and answers.

Models reach the platform two ways. A customer points Calvin Risk at a company-internal endpoint or uploads model parameters as an MLFlow file, working through the Calvin API or the app. The vendor names Azure AI Studio, MLFlow, Sagemaker, TensorFlow and PyTorch among the tools the platform integrates with, and its micro service architecture supports on-premises, virtual private cloud and Calvin-hosted installation. [s2, s4, s3]

CB Insights places Calvin Risk in a competitor set it heads with Naaia, and its comparison list adds Modulos, Holistic AI, Enzai and Asenion. CB Insights describes each of them as an AI governance, risk or compliance platform, which is the layer of Calvin Risk's product that follows published regulation.

The money step is what those descriptions do not mention. Calvin Risk reports a value at risk in currency for each business use case, combining a risk frequency with a severity estimate whose risk causes and effects come from ISO 31000. A buyer comparing platforms can ask each vendor for that number and see who produces one.

Risk.net judged the product in a different category again, naming Calvin Risk model risk management product of the year in its 2024 Markets Technology Awards. That is a model validation category rather than an AI security one, which tells a buyer which shortlist the product has been ranked on. [s15, s2, s3, s14]

Four organizations appear by name in the reviewed sources, each in a different kind of engagement. Aviva and Lloyds Banking Group appear as software-development partners, Lufthansa Industry Solutions as the subject of an evaluation, and SBB CFF FFS in a workshop testimonial.

The Aviva relationship runs through the insurer's innovation team. The joint announcement frames it as exploring how customisable software can help insurers assess and manage the technical, ethical and regulatory risks of using AI models, and a Forbes contributor article says Calvin Risk has been developing its software in partnership with Aviva and Lloyds Banking Group.

The Lufthansa work is the part an outside party corroborates. Lufthansa Industry Solutions' own newsroom states that it had its AI Maintenance Assistant evaluated by TÜV SÜD and Calvin Risk against key parts of the EU AI Act, and the operational tool for those reviews is one Calvin Risk built with TÜV SÜD.

No revenue figure appears in the reviewed sources, so a buyer sizing the business has awards and named engagements to go on and nothing else. Risk.net named the product model risk management product of the year in 2024, and Calvin Risk reports further awards from Swiss InsurTech Hub and Hub France IA. [s1, s10, s13, s17, s11, s14, s8]

Two of the three co-founders run the company. Julian Riebartsch is CEO, with a computational physics degree from ETH Zurich and four years in venture capital, most recently at btov Partners. Syang Zhou is CTO, with six years in risk management and technology consulting and a place in ETH's stochastic finance research group.

The academic tie is documented on both sides. ETH Zurich names Calvin Risk AG among the 26 spin-offs it created in 2022, and Calvin Risk says the statistical methodology behind its quality metrics was co-developed with ETH Zurich over more than ten person-years of work by its data science team and founders.

Neither founder profile names a company previously built or sold. No prior exit appears on the about page, in the Forbes article or in the CB Insights profile, so a buyer weighing the team has in-domain roles and the ETH connection to go on rather than a track record of shipped companies. Calvin Risk was hiring one forward deployed engineer in Zurich at the time of writing. [s6, s16, s3, s13, s15]

No attestation Calvin Risk itself holds appears on any surface probed on 21 August 2026. The trust and security subdomains do not resolve, and neither does a nonsense control subdomain, so the non-resolution is meaningful rather than a DNS artifact. The /trust, /security and /compliance paths return 404, and a web search for a certification returned none.

The compliance the platform carries belongs to its customers. The EU AI Act module classifies a customer's system as high risk or not, stores the evidence that classification requires, and the severity model takes its risk causes and effects from ISO 31000.

The record's security-relevant detail is about deployment. Calvin Risk's micro service architecture supports on-premises, virtual private cloud and Calvin-hosted installation, so a regulated buyer can choose where the assessment runs. [s19, s21, s3, s2]

Add analyzed competitors to compare them side by side with Calvin Risk.

Calvin Risk sells to regulated enterprises that run enough AI to need a register of it. The industries the reviewed sources reach are insurance, banking and transportation, and a Forbes contributor article adds pharmaceuticals and telecoms as newer ground.

No security title appears among the named contacts in the reviewed sources. Aviva's named contact is its chief innovation officer, SBB's is an AI adoption lead, and Lufthansa Industry Solutions' is a director of artificial intelligence. Calvin Risk's own seed announcement names the Audit Research Center as a partner for third-line defence solutions and describes the platform serving control functions and management.

Risk.net's award points the same way. It judged Calvin Risk in model risk management, a risk-function category, which places the product among risk tooling.

The measurement layer is where Calvin Risk's engineering concentrates. The company says it builds its AI quality metrics from proprietary industry data and a statistical methodology co-developed with ETH Zurich, and that its data science team and founders have put more than ten person-years into the metrics and the platform.

The LLM side is the more specific claim. Calvin Risk states that its LLM metrics come from 150 million tokens of questions, retrievals and answers, and that the module computes 6 performance, 11 robustness and 2 fairness metrics.

The severity model is the part that turns a score into money. Calvin Risk selects risk causes and effects using ISO 31000 and combines model risk frequencies with severity to report a value at risk in euros or dollars for each business use case. Customers can add their own metrics to the platform alongside the shipped ones.

Partnerships and programmes are the routes into accounts the record documents. The Aviva work began through the insurer's innovation team, the Lufthansa Industry Solutions work came through a joint evaluation with TÜV SÜD, and the seed announcement names the House of Insurtech Switzerland innovation garage among its routes.

Recognition is a recurring feature of the company's own announcements. Risk.net named the product model risk management product of the year in 2024, and the company reports being named top AI risk management tool for banking by Hub France IA and InsurTech of the Year 2024 by Swiss InsurTech Hub.

Europe is the stated reach. Calvin Risk's seed announcement says the round will expand its market reach across Europe, and a Forbes contributor article says the company is beginning to work with customers in pharmaceuticals and telecoms beyond its financial services start.

Calvin Risk publishes no price. Its pricing page says the platform is available to companies under an annual site license and asks a prospect to contact the company for detail.

The evaluation path is a qualifying one rather than a self-serve trial. The same page says a prospect can qualify for a test validation using the Calvin-hosted public cloud, which puts Calvin Risk between a prospect and a hands-on look at the product.

The pricing page leaves a prospect without a number. It names an annual site licence and no figure, so cost enters the evaluation only after a conversation with the company.

Onboarding is the customer's work. Calvin Risk takes models over a company-internal endpoint or an MLFlow upload, and assessments run through the Calvin API or the app.

Hosting is the customer's choice. The micro service architecture supports on-premises, virtual private cloud and Calvin-hosted installation.

Calvin Risk also lends its own judgment in named engagements. Lufthansa Industry Solutions' newsroom names Calvin Risk alongside TÜV SÜD as the evaluators of its AI Maintenance Assistant, and Greater Zurich Area's news item records that the two of them found key EU AI Act requirements met.

No attestation Calvin Risk itself holds appears on any surface probed on 21 August 2026. The trust and security subdomains do not resolve, a nonsense control subdomain does not resolve either, the /trust, /security and /compliance paths return 404, and a web search for a certification returned none.

What the record does carry is a third party's use of a tool built with the company. TÜV SÜD contributed its own AI quality framework to the Lufthansa evaluation, and the operational software for those digital reviews is a tool Calvin Risk built with TÜV SÜD inside that framework.

The OECD's Catalogue of Tools and Metrics for Trustworthy AI has listed the tool since October 2023. The catalogue states that its entries are solely those of the originating authors and are not vetted or endorsed by the OECD, so the listing is a directory record rather than an assessment. The entry describes Calvin Risk's primary focus as second line-of-defense measures.

Calvin Risk integrates into the customer's existing machine learning stack rather than asking for a new one. The vendor names Azure AI Studio, MLFlow, Sagemaker, TensorFlow and PyTorch among the tools the platform works with, and onboards systems through endpoint integration for retrieval and LLM applications or an MLFlow upload for tabular models.

The extension point is the metric library. Customers can add their own quality metrics and scores to the platform, which puts a customer's internal validation standards inside the same reports as the shipped ones.

The one partner-built surface in the record belongs to TÜV SÜD. The Evidence Management Tool, QLAIM, is a joint build. It works within the AI quality framework TÜV SÜD defines rather than within Calvin Risk's own, so a buyer reading assurance off that arrangement is relying on TÜV SÜD's framework and Calvin Risk's tooling together.

Two of the three co-founders the about page names run the company. Julian Riebartsch is CEO with an ETH Zurich computational physics degree and four years in venture capital at btov Partners, and Syang Zhou is CTO with six years in risk management and technology consulting and a place in ETH's stochastic finance research group.

The research connection is the team's distinguishing credential. ETH Zurich names Calvin Risk AG among the 26 spin-offs it created in 2022, and the company says its metrics methodology was co-developed with ETH Zurich.

The gap in the record is a delivery track record. No prior company built or sold appears in the founder profiles, the Forbes article or the CB Insights entry, so a buyer weighing execution risk has domain roles and an academic pedigree to go on rather than a shipped precedent.