How We Got Here (Cloud Security → CNAPP)

We watched posture tooling commoditize, driving ASPs down and pushing vendors to deliver outcomes. A (very) condensed arc:

• 2012–2014 – Chaos/“Simian Army” era; early posture watchers emerge.
  
  
• 2015–2018 – CSPM & CWPP formalize; acquisitions consolidate posture tooling.
  
  
• 2016–2019 – Benchmarks (CIS) and native cloud CSPM features arrive; major misconfig incidents expose identity/posture gaps.
  
  
• 2020–2021 – Shift‑left/IaC scanning and CIEM rise; CNAPP is coined to unify controls.
  
  
• 2021–2023 – Agentless discovery and attack‑path context popularized; consolidation accelerates.
  
  
• 2022–2025 – CNAPP becomes the default control plane that blends CSPM, CWPP, CIEM and IaC.
  
  

Lesson: Visibility matures into integrated prevention.

With the cloud storing an increasing share of data, more data security controls have emerged, echoing the earlier evolution of cloud security. For example, many CNAPP providers now offer some kind of data security posture management (DSPM) capability. Yet these only provide partial protection for two reasons: 1) most have not yet progressed beyond visibility to include prevention and 2) more importantly, data is everywhere and CNAPP-bundled DSPM focuses on cloud IaaS only, leaving critical gaps for data on endpoints or on-premises servers running mission-critical systems, as is typical in large, heavily-regulated industries.

Why DSPM Visibility Isn’t Enough (Especially in the AI Era)

• Data is inherently contextual and diverse. “Sensitive” varies by company, team, and workflow.
  
  
• AI fragments data. We no longer only share files–we share fragments: prompts, snippets, embeddings, screenshots, and summaries. These fragments move across endpoints, SaaS, and LLMs.
• Labels break down. In a fragmented world, labels/tags applied to data at rest rarely follow every derivative or snippet. Result: exfiltration despite accurate classification.
  
  

Implication: DSPM must evolve from labeling-at-rest to tracking-and-protecting-in-motion.

Data flowing in and out of AI is critical and worth protecting. In fact, it’s the data coming out of AI that is more difficult to track and control, yet organizations increasingly rely on it to make decisions. Worse yet, these fragments cannot be labeled in conventional ways, meaning that a large and growing share of data is beyond the reach of traditional data security controls. 

Modern DSPM must go beyond data-at-rest labeling to prioritize and auto-remediate risks proactively to shrink the attack surface, while tracking and protecting data in motion to prevent data exfiltration.

What “Reimagined Data Protection & Insider Risk” Requires

Foundational capabilities (platform, not point products):

1. Data at Rest Discovery across endpoints, servers, cloud; plus durable classification.
2. Data Access Minimization to reduce blast radius.
3. Cross‑Surface Data Lineage to map how sensitive data and its fragments move between users, apps, and surfaces.
4. Derivative/Knowledge Loss Protection to stop snippet‑level fragmented data exfiltration (not just file egress).
5. Human Insider Risk detection and Non‑Human/Agent Risk controls for automated actors.
6. Real‑Time Coaching & Awareness in the flow of work to reduce risky behavior without blocking productivity.

Lineage Over Labels – and Why Endpoints Matter

Lineage connects sources → derivatives → destinations so controls trigger on meaning, not just file names or static tags. To build reliable lineage:

• High‑fidelity endpoint telemetry is non‑negotiable. It’s how you observe creation, copy/paste, screenshots, generative use, and app‑to‑app flows.
  
  
• Scale and stability are hard‑won. Lightweight agents that perform across diverse OS/app stacks take years of iteration.
  
  

Bottom line: Without mature endpoint tech, lineage – and therefore effective protection in an AI world – is incomplete.

Engineering Reality Check

Endpoint excellence is the steep part of the curve and the approach that most effectively protects data.

Why? Endpoint controls provide the most comprehensive view of user intent and actions right when data interactions occur. Therefore, they provide the most context, which is essential for effective controls.

What Pure‑Play DSPM Vendors Are Trying Now

• Build or buy endpoint controls/data lineage. Powerful, if you have the DNA and patience to mature it.
  
  
• “Helper products” for legacy DLP. AI‑assisted classification to tune other vendors’ DLP. 

Building a mature, scalable endpoint agent with lineage tracking is hard. Agents take years of iteration, while API connectors are easy by comparison. Even after decades of development, most legacy DLP endpoint agents still require the toxic tradeoff between security and usability.

Furthermore, these “helper products” may sometimes help at the margins, but the gains are incremental and don’t solve for fragmented data visibility or false negatives at the source. As a result, “garbage in, garbage out” problems persist. They also don’t fix the performance issues that plagued legacy tools and frustrate users. As more unlabeled data is created, consumed, and shared, the gaps stay open and grow, increasing security debt. The result is the illusion of coverage while material exposures remain. In practice, these “helper products” can hurt more than help by creating false confidence, like painting over rot instead of repairing the structure.

Risk: Commoditization → pricing pressure → scramble for a second act without the core capability (endpoint‑anchored lineage).

What the Market Actually Needs

A unified platform that:

• Starts with proven endpoint DLP + lineage capturing fragment flow across people, apps, and agents.
  
  
• Integrates discovery → posture → protection → coaching in one loop.
  
  
• Covers data at rest and in motion equally well – across endpoints, SaaS, cloud, and AI.
  
  

Capability Snapshot

Illustrative: vendor capabilities vary.

Buyer’s Data Protection Checklist for the AI Era

• Can you prove lineage from source → derivative → destination across copy/paste, screenshots, LLM prompts/outputs?
  
  
• Do you protect snippets and derived knowledge in fragments, not just labeled files?
  
  
• Can the platform minimize access and fix posture, not just alert on it?
  
  
• Does the platform provide insights into insider risk and real-time end-user coaching?
  
  
• Is coaching integrated and results in measurable risk reduction?

Conclusion

History doesn’t repeat, but it rhymes. Like CSPM before it, DSPM’s first act – visibility – will commoditize. The durable second act is lineage‑driven protection anchored on mature endpoint technology, wrapped in a platform that unifies discovery, posture, prevention, and coaching. If you’re maturing AI and data security, pick protection that tracks meaning, not just labels.