TL;DR: Legacy DLP built on regex policies is failing to keep pace with AI-driven data movement across endpoints, browsers, SaaS, and unauthorized apps, according to Orion’s event recap. The control gap is no longer classification first or more alerts, but context-aware detection that can see where data actually goes and why.
At a glance
What this is: Orion argues that AI-driven data movement has outgrown legacy DLP playbooks built on regex policies, classification-first workflows, and fragmented point controls.
Why it matters: Security and IAM teams need data movement visibility because AI apps and shadow workflows can move sensitive information faster than static policy stacks can classify or contain it.
Context
Data loss prevention is meant to identify and stop sensitive information from leaving approved boundaries. In practice, the control has often been split across endpoint, email, network, SaaS, and now AI tools, which creates blind spots when data moves across channels faster than policy can follow. That becomes an identity and governance issue as soon as employees can use approved or unauthorized AI apps with access to internal data.
Orion’s event recap frames a broader shift in data security governance: the question is no longer whether a label exists, but whether the organisation can observe data in motion and act before exposure becomes routine. The article’s examples show why classification-first programmes struggle when AI adoption outpaces policy design and why context, not just content matching, is now central to control effectiveness.
Key questions
Q: What breaks when DLP relies only on regex and keyword scanning for AI use?
A: Regex and keyword scanning miss the decision point that matters most, which is the origin and sensitivity of the data before it is pasted. They can overblock harmless content and still miss confidential code, customer records, or strategy documents. In AI workflows, that creates both false confidence and weak enforcement, because the control sees text patterns but not business context.
Q: Why does AI-driven data movement increase exposure even when users have normal access?
A: Normal user access can still produce abnormal risk when AI tools and connected apps can copy, transform, and route data into places the security team does not monitor well. The issue is not only privilege, but unobserved pathways. That is why context-aware detection matters more than another layer of static policy.
Q: What are the signs that a DLP programme is not working as intended?
A: Common warning signs include high false-positive volume, noisy alerts that security teams ignore, weak coverage across SaaS and endpoints, and enforcement that blocks routine work more often than risky behaviour. If the system cannot distinguish sensitive data from normal usage, or cannot show how data moved, the programme is too brittle to support reliable response or compliance evidence.
Q: Should organisations prioritise data movement visibility before full data classification?
A: Yes, when AI adoption is moving faster than taxonomy work. Full classification is valuable, but it should not block immediate protection. If you can already see where data is going, you can reduce exposure now and feed those observations back into labelling, policy, and governance work later.
Technical breakdown
Why regex-based DLP misses AI-era data movement
Legacy DLP typically relies on pattern matching, dictionaries, and static policy rules to spot sensitive content. That model works poorly when users move data through AI assistants, browser-based tools, desktop copilots, and connected SaaS apps because the risk is not just the content, but the path, timing, and follow-on action. The article’s core point is that false positives, missed threats, and fragmented queues are a control failure, not just an operations problem. When the same file can be copied, transformed, zipped, renamed, and re-uploaded across multiple surfaces, regex alone cannot reconstruct intent or sequence.
Practical implication: teams need controls that correlate movement across surfaces instead of treating each alert as an isolated event.
How context-aware detection changes the data security model
Context-aware detection looks at surrounding signals such as source, destination, sequence, user context, and behavioural anomalies. In the article’s framing, that means understanding where data is going, not only what the data contains. This is closer to behavioural monitoring than to classic content inspection. It also creates a governance bridge to identity because the system has to interpret who is moving data, from which endpoint, into which app, and whether that pattern matches the organisation’s normal operating model. In modern environments, the control boundary is the data flow itself, not the label on the file.
Practical implication: establish telemetry that joins endpoint, browser, SaaS, and AI activity into one detection view.
Why classification-first programmes delay protection
Classification remains useful, but it becomes a bottleneck when organisations insist on fully labelling data before they can protect it. The article describes a shift toward classifying data in motion so detections can feed back into compliance and labelling workflows later. That reverses the old sequence. In an AI-enabled environment, waiting for perfect taxonomies can mean waiting until exposure is already occurring. The governance lesson is that operational protection and data governance must be coupled, not staged as separate projects with long delays in between.
Practical implication: deploy in-motion detection first, then use the findings to accelerate labelling and policy refinement.
Threat narrative
Attacker objective: The objective is to exfiltrate or misuse sensitive company data through legitimate-looking user activity and unseen AI-connected workflows.
- Entry occurs when employees use AI tools, browser copilots, desktop apps, or unauthorized applications that connect to internal data sources.
- Credential or access abuse follows when those tools operate with ordinary user privileges but can still reach sensitive content and movement paths the security team has not seen.
- Escalation happens as data is copied, transformed, zipped, renamed, and moved across multiple channels, making isolated controls miss the full sequence.
- Impact is unauthorized data exposure, with the organisation losing visibility into where sensitive information went and why.
Breaches seen in the wild
- Sisense breach 2024: A credential in Sisense's GitLab reportedly opened S3 buckets of customer tokens, passwords and certificates; CISA urged a full reset.
- CISA Private-CISA GitHub leak 2026: A CISA contractor's public GitHub repo exposed AWS GovCloud admin keys, Artifactory credentials and plaintext passwords for six months.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
AI-driven data movement is now a governance problem, not just a tooling problem: once employees can route sensitive information through copilots, browsers, and shadow applications, the control question shifts from simple inspection to accountable visibility. The old DLP stack assumed data would pass through predictable channels and stable policies. That assumption is now broken, so programme owners need to treat data movement as a governed workflow, not a series of disconnected alerts.
Pre-classification is no longer a defensible prerequisite for protection: organisations that insist on complete labelling before deploying effective controls are choosing process over exposure reduction. The article shows the better sequence is to detect data in motion first, then backfill classification and compliance workflows from real activity. Practitioners should read that as a lesson in control ordering: protection must start where the risk is happening, not where the taxonomy is complete.
Context is the missing control plane in modern data security: a single download or upload is rarely meaningful on its own, but the surrounding sequence can reveal misuse, exfiltration, or policy drift. That makes data security increasingly dependent on correlated telemetry across endpoint, browser, SaaS, and AI surfaces. For security leaders, the lesson is to govern the path of data with the same seriousness they apply to the data itself.
Fragmented DLP creates fragmented accountability: when endpoint, email, AI, and SaaS teams all own separate queues, no one owns the end-to-end risk story. That is why the article’s boardroom example matters: once leaders can show where data is going, the conversation changes from noisy alert handling to measurable exposure reduction. Practitioners should align detection, compliance, and business stakeholders around the same movement evidence.
From our research library:
- Business leaders plan to spend $124 million on average on AI in 2026, and 91% say data security and risk will shape their AI strategy.
What this signals
AI-driven data movement is exposing the limits of legacy DLP operating models: the real issue is no longer whether a policy can match a string, but whether the programme can reconstruct how data moved across connected tools. That pushes security teams toward sequence-aware telemetry and away from surface-by-surface ownership models.
Data security teams should treat AI adoption as an access-path expansion problem: once employees can use unapproved or lightly governed tools to reach internal data, the programme needs visibility into destinations, transformations, and handoffs. The practical shift is from blocking everything equally to identifying which flows deserve intervention.
Control ordering now matters as much as control coverage: if protection waits on perfect classification, the programme will always lag the business. The more durable model is to protect observed flows first and let classification, compliance, and governance catch up from live evidence.
For practitioners
- Map current data movement across AI and SaaS surfaces Instrument endpoint, browser, SaaS, and AI-connected workflows so you can see where sensitive data actually travels before it is classified or blocked.
- Shift from pattern-only rules to correlated detections Use behavioural and sequence-based signals to connect download, transform, rename, and upload events into one case instead of treating them as separate alerts.
- Deploy in-motion protection before full taxonomy completion Allow controls to act on observed data flows while labelling and policy work continues, rather than waiting for a perfect classification model.
- Unify security and business context for escalation Create a shared workflow with functions such as HR or legal so elevated-risk activity is judged with context, not blanket assumptions.
Key takeaways
- Legacy DLP struggles because AI tools change both the path and the pace of sensitive data movement.
- Fragmented endpoint, email, SaaS, and AI controls cannot explain a full exposure chain on their own.
- Security leaders need in-motion visibility first, then classification and policy refinement can follow.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK addresses the attack and risk surface, while NIST CSF 2.0, CIS Controls v8 and NIST SP 800-53 Rev 5 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.DS-01 — Data-at-rest protections | The article centers on protecting sensitive data as it moves across modern environments. |
| DE.CM-06 — External Service Provider Activities Monitored | AI apps and SaaS channels extend data handling beyond traditional perimeters. | |
| Recommendation — Apply PR.DS-01 to protect sensitive data flows with controls that follow the data, not just the storage location. Use DE.CM-06 to monitor third-party and external service data paths that can move sensitive information. | ||
| CIS Controls v8 | CIS-13 — Data Protection | The article is fundamentally about improving protection over sensitive data movement. |
| Recommendation — Implement CIS-13 to reduce exposure by combining detection, classification, and policy enforcement around data use. | ||
| NIST SP 800-53 Rev 5 | AU-2 — Event Logging | Correlated detections across endpoint, browser, SaaS, and AI surfaces depend on usable telemetry. |
| Recommendation — Configure AU-2 to capture the events needed for end-to-end data movement correlation and investigation. | ||
| MITRE ATT&CK | TA0009;TA0010 — Collection; Exfiltration | The threat pattern described is collection and movement of sensitive data through legitimate-looking channels. |
| Recommendation — Map observed data movement to TA0009 and TA0010 to prioritise detection of collection-to-exfiltration chains. | ||
Key terms
- Context-Aware Detection: Context-aware detection is a method that evaluates surrounding meaning before deciding whether data is sensitive. Instead of relying only on patterns, it uses signals from nearby text, file content, or communication context to reduce misclassification. This approach is especially useful in DLP, where isolated terms can look risky but are actually benign.
- Data Movement Visibility: Data movement visibility is the ability to see where sensitive information appears as it travels across applications, users, and workflows. It goes beyond pattern matching in a single file or message and shows how data spreads across SaaS, cloud storage, collaboration tools, support systems, and AI tools in practice.
- Classification-First DLP: Classification-first DLP is a programme design that requires data to be labelled before protection can be applied or tuned. It can improve governance when complete, but it becomes a bottleneck when AI adoption changes the data path faster than taxonomies and policies can be built.
- Fragmented Signal: Fragmented signal is the loss of detection quality that happens when separate tools each see only a small part of a data event. The result is noise, missed incidents, and weak investigation context because no single control plane reconstructs the full sequence.
Deepen your knowledge
The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management. It helps security and identity practitioners build the control thinking needed for modern access and exposure problems.
Published by the NHIMG editorial team on August 14, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org