Data Security Platforms

Forrester's current evaluation of this market identifies AI and quantum security as the forces shaping what these platforms offer.

Quantum security is an unusual thing to find in a buying decision made this year, because the machines that would break current encryption do not exist yet. In most security categories that would make it a research topic rather than a criterion.

It belongs here because data security is the one discipline where the clock runs against the defender.

An endpoint compromised today is remediated and returns to service. A network intrusion is contained. But data exfiltrated today is exposed permanently, and data encrypted today with algorithms that may be broken in a decade is a decision being made now about a risk that arrives later.

That asymmetry explains most of what is distinctive about this category.

What a data security platform consolidates

A DSP brings together a set of controls that were historically bought separately and rarely worked together.

Data discovery and classification, meaning finding sensitive data across the estate and labelling what it is. Data loss prevention, controlling where it can go. Encryption and key management. Tokenisation and masking, replacing sensitive values with substitutes for use in environments that should not hold the real thing. Information rights management, keeping controls attached to a file after it leaves your systems. Privacy tooling for subject access, consent, and retention. And threat and risk visibility over who is accessing what.

The criteria in Forrester's evaluations track that scope. Microsoft's maximum scores in the Q1 2023 edition covered data classification, data threat and risk visibility, masking and redaction, encryption, rights management, privacy use cases, and integrations for zero trust. Forcepoint's in the Q1 2025 edition covered classification, data loss prevention, information rights management, and tokenisation.

Classification appears in both, and it is the foundation everything else rests on.

The twenty-year failure

Data classification has been the most persistently unsuccessful discipline in enterprise security, and the reason is that for two decades it depended on people.

The model was labelling. Users would mark documents as confidential, internal, or public at creation, and downstream controls would enforce policy based on those labels.

It never worked at scale. Users do not know the classification scheme, do not remember it, and have no incentive to apply it accurately. A label that restricts sharing makes their job harder, so the rational choice is the least restrictive option or whatever the default is. Meanwhile the volume of unstructured data grew faster than any labelling effort, and the majority of sensitive information ended up in files nobody classified, in systems nobody inventoried.

Every control downstream inherits that failure. Data loss prevention cannot prevent loss of data it cannot identify. Rights management cannot protect what was never marked. Access reviews cannot flag over-permissioned access to content of unknown sensitivity.

Machine classification changes this materially, and it is the reason this category became viable as a platform rather than a collection of tools. A system that reads content and determines sensitivity from what the data actually is, rather than from what someone labelled it, removes the human dependency that broke the model.

Forrester's assessment of Varonis in the current evaluation reflects where the value lands: customers praised the careful, planned use of AI and automation specifically in workflow and data classification, and Forrester credited a vision of using deep data insight to automate remediation.

Automated remediation is the point. Classification that produces a report tells you how large the problem is. Classification connected to action fixes over-permissioned files without a human reviewing each one.

Inside the evaluations

The Forrester Wave: Data Security Platforms, Q1 2023 named Microsoft a Leader with maximum scores across the criteria listed above, and Google Cloud a Leader with the highest current offering score of any vendor evaluated.

The Forrester Wave: Data Security Platforms, Q1 2025 named Varonis a Leader with the highest score in both current offering and strategy, more maximum criterion scores than any other vendor, and recognition as a Customer Favorite. Forrester's characterisation is of a data security veteran carrying expertise built in on-premises environments into cloud ones.

Forcepoint placed as a Strong Performer, scoring among the top three in current offering, and was positioned by Forrester for organisations needing mature data loss prevention and data controls, particularly with a data-centric zero trust focus.

A single vendor taking the top score in both dimensions plus the customer marker is a dominant result, and it is worth noting what kind of vendor did it. Varonis is a specialist in a category where the two Leaders from the previous edition were hyperscale platform vendors.

The conditional that decides the suite question

Heidi Shey's assessment of Microsoft in the 2023 evaluation contains the most useful seven words in either report: "Microsoft shines with its ecosystem approach, if you go all in".

That conditional is the entire suite-versus-specialist decision, stated honestly.

A platform vendor's data security capability is excellent inside its own estate, because it has native visibility into the systems, the identities, and the content. It is weaker outside, because everything beyond the estate requires integration that competes for roadmap attention against native capability.

Most enterprises are not all in on anything. They run one vendor's productivity suite, another's infrastructure, a third's data warehouse, and dozens of SaaS applications procured departmentally. Data sits in all of them.

Which is the structural argument for the specialists. A vendor whose entire business is data security across heterogeneous environments has no incentive to be better in one estate than another, and no adjacent product line pulling its attention.

The counter-argument is depth of integration. Native access to a platform's internals is genuinely better than API access from outside, and organisations that really are concentrated in one ecosystem give up something by choosing a specialist.

The honest test is to inventory where your sensitive data actually lives, by volume and by sensitivity, before evaluating anything. That inventory decides the question, and most organisations discover the distribution is wider than they assumed.

Generative AI opened a new door

Forcepoint's framing of the current situation names the problem plainly: sensitive data is surging into generative AI systems alongside multi-cloud environments, on-premises servers, endpoints, email, and collaboration applications.

The mechanism is worth being precise about, because there are several distinct exposures and they need different controls.

An employee pasting confidential material into a consumer AI tool is a classic data loss prevention problem with a new destination. It is detectable by the same controls that watch uploads and web traffic, provided the policy covers it.

An organisation building retrieval systems over its own content is a different exposure. The content stays inside, but the permission model changes. If retrieval surfaces passages from documents a user cannot open, the information leaks through the answer rather than through the file. That is not a DLP problem, it is an access control problem appearing in a new place.

Fine-tuning on sensitive data creates a third. Information incorporated into model weights cannot be deleted the way a record can, which sits awkwardly with retention obligations and subject access rights.

And agents introduce a fourth, where software with broad access acts on data at machine speed, and the audit question becomes what an agent read as well as what it changed.

A data security platform that treats generative AI as one more egress channel is solving the first problem only. The vendors worth taking seriously distinguish between them.

Harvest now, decrypt later

The quantum criterion is the clearest illustration of why this category thinks in decades.

The concern is not that encryption breaks tomorrow. It is that encrypted data intercepted today can be stored and decrypted whenever the capability arrives. Anything with a long confidentiality requirement, government records, health information, intellectual property, financial and legal records, is exposed to a decision being made now.

Standards bodies have published post-quantum algorithms and migration guidance exists, but enterprise cryptography is embedded in applications, protocols, hardware modules, and certificates accumulated over decades, most of which nobody has inventoried.

Which produces a practical requirement that sounds mundane and is not: knowing where cryptography is used, by which algorithm, protecting what, with what key lifetime. That inventory is a prerequisite for any migration, it takes years to build, and platforms that maintain it are producing something with genuine value regardless of when quantum computing arrives.

For most organisations the reasonable position is to establish crypto-agility, meaning the ability to change algorithms without rebuilding applications, and to prioritise data whose confidentiality requirement extends past the horizon where the threat becomes plausible.

What the category is actually asking

Data security differs from every adjacent security discipline in one respect. Network security, endpoint security, and application security defend systems, and systems get replaced. Data outlives the systems that hold it, the people who created it, and frequently the organisation's memory of why it exists.

That produces the discipline's real question, which is not which controls to buy but what you actually hold. Most enterprises cannot answer it. They know their systems, their applications, and their users, and they have an approximate and optimistic view of their data.

Every capability in this category depends on closing that gap. Classification is how you find out what you have. Everything else is what you do about it.

The organisations that get value here start with discovery and accept what it tells them, which is usually that sensitive data exists in more places, in more copies, accessible to more people, than anyone believed. The ones that do not end up with a platform enforcing sophisticated policy over the portion of their data they already knew about.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.