Murex

MX.3 has been a complete cross asset trading, risk and post trade platform since 1986, and without its neural networks it still prices, books, clears and settles, only slower and on more hardware. The vendor itself pitches machine learning as a way to run costly models without prohibitive infrastructure, which makes it acceleration, and its July 2026 research initiative is framed as long term exploration.

The neural network approximates a model whose exact answer the vendor already has, so a comparison control is architecturally available, yet none is described. No tolerance threshold, fallback to the exact model or review of a surrogate priced number before it reaches a book appears.

A published white paper sets out the neural network architecture and training needed for trading grade precision, the head of quantitative research is on record, and the research arm publishes openly. Because the network replicates a known exact model, fidelity is measurable against a true answer, yet no error bounds, drift monitoring or external validation are published.

DZ Bank, Rabobank and Julius Baer head a deep named client library, and analyst recognition is current, including a 2026 category leader placement for enterprise market risk. No client story concerns the machine learning, and the 7x speed and 4x energy figures come from a chip maker's benchmarks in the vendor's own lab.

Models are trained centrally on the vendor's infrastructure and shipped to many clients for inference, using derivative contract data, yield curves and volatility surfaces. Whether any training data comes from client books, or one client's positions shape a model another runs, is described as engineering and never committed to.

Only a standard privacy policy exists, with nothing addressed to the platform. Under the managed services, where the vendor operates the environment, what its staff can access and how long data is retained matter more than under a license, and the site does not distinguish the two.

The security page names mechanisms: fine grained role authorization with multi factor sign on, a four eyes principle on sensitive changes, encryption at rest and in transit, static analysis and penetration testing before release. The one credential, a SOC 2 Type I attestation correctly scoped to the managed service, tests design at a point in time, and no ISO 27001 appears.

The company licenses MX.3 for clients to run and also performs processing for them through its managed services, the side where supervision would attach if anywhere. No examination, registration or supervisory relationship is claimed for either.

A neural network approximating a derivative valuation has no protected class or individual subject, so this gap matters less than it would in credit decisioning. The research program does list explainable and responsible AI and automation across capital markets, an intention toward decision support with no governance position behind it.

A surrogate that drifts from the exact model produces a wrong valuation that flows from pricing into accounting and regulatory capital. For output entering financial statements a SOC 1 report would be the relevant assurance, yet only a SOC 2 Type I is published, with no allocation of the loss.

The pricing models are the vendor's own, and the architecture and training approach are published, which tells a buyer more than a supplier name would. Cloud and hardware partners are infrastructure rather than model suppliers, and no generative capability appears to ship.

MX.3 is the system of record for trading, risk and post trade, with a tiered service oriented architecture and a documented REST layer. Published envelopes, 1,000 imported trades per second at peak and margin across 20,000 collateral agreements in ten minutes, turn integration claims into testable ones.

On premises license, cloud, vendor managed service, upgrade as a service and managed business process delivery are each named, with Amazon Web Services under at least one client and a secure deployment guide for both settings. No residency statement or regional hosting map covers a platform operating worldwide, though one Brazilian case notes local requirements met.

Pricing is available only through sales. The managed business process service is described as pay as you go with no added infrastructure, aimed at smaller banks once priced out of trade pricing and valuation adjustments, a model and a target buyer without a price.

About 300 institutions, 60,000 daily users and 60 countries are stated, across six named segments from banks and clearing houses to energy firms and public agencies. MX.3 underpins a swap clearing platform carrying roughly 90 percent of over the counter vanilla swap volume and processes over 35 million foreign exchange cash trades a day.