VAST Data's DataEnclave lets firms run AI on the data they keep locked away

The product addresses a contradiction described in that report: businesses are spending heavily on artificial intelligence while the data most likely to generate commercial value from it remains locked away.

Every company holds data it guards more closely than the rest, the report notes. For a bank that may be millions of customer transactions; for a pharmaceutical company, years of proprietary research; for a cybersecurity business, detailed information about threats and vulnerabilities across its networks. Putting sensitive customer information, intellectual property or regulated data into an external AI service raises questions about who can access it, where it is processed and what happens to it afterwards. For companies subject to GDPR, AI and industry regulation or national security requirements, the report says, getting those questions wrong can carry serious consequences.

Much of the recent discussion around enterprise AI has centred on the models themselves, which are described as the smartest, fastest or cheapest. The model, however, is only part of the equation. A bank trying to spot sophisticated fraud becomes more effective the more context an AI system has across transaction histories, customer behaviour and previous incidents. A pharmaceutical company sitting on years of experimental data could combine it with powerful models to find solutions that would otherwise take months to uncover. The opportunity described is not to improve the AI but to put information a business already owns to work in ways that improve products, reduce risk, speed up research or create new services.

The difficulty arrives when that data actually has to be used rather than stored. An AI model has to process information to analyse it, which traditionally creates a point at which sensitive information may become visible inside the infrastructure running the workload. Confidential computing is designed to address this by using hardware-backed protections to isolate data while it is being processed, rather than relying only on rules governing who should have access. VAST Data describes the result as a locked room for computation: the data and the AI model can enter, but the infrastructure operator, and sometimes even the model provider, does not get access.

VAST Data builds a unified AI operating system that combines data storage, databases and compute infrastructure on a single platform. DataEnclave reverses an assumption behind many AI deployments. Instead of asking a business to move sensitive information somewhere an AI provider can reach it, the model is brought to the information.

Among the uses cited are a financial services company analysing sensitive customer information without handing the underlying records to its AI model, a cybersecurity business running AI against highly confidential threat data, and a government applying AI capabilities to information it would never consider placing into a public AI service. Startups may gain as well. Europe has built businesses holding highly specialised datasets in finance, healthcare, defence and industrial technology, and much of their competitive advantage lies in information accumulated over years. If that data can safely be combined with a wider choice of AI models, those businesses would no longer have to make the same trade-off between access to the latest AI capabilities and protection of their own intellectual property.

VAST Data's technology does not remove the need for governance, according to the report. Companies still have to decide which data should be used, which models they trust and what employees or AI agents should be allowed to do with the results. Confidential computing does not make those decisions disappear, but it could remove one barrier. Sifted writes that for many businesses the next significant opportunity may come less from another jump in model performance than from finally being able to use information they have kept closed off.