Notifications
Clear all
Topic starter
14/08/2026 6:35 pm
TL;DR: AI agents can move data at machine speed, through API calls and model requests that legacy DLP was never built to see, according to Orion. The governance challenge is shifting from reviewing human behaviour to enforcing action-level controls on software that can act hundreds of times per session.
NHIMG editorial — based on content published by Orion: DLP for AI Agents
By the numbers:
- Gartner expects 40% of enterprise applications to ship with task-specific AI agents by the end of 2026, up from less than 5% in 2025.
- Gartner expects 15% of day-to-day work decisions to be made autonomously by agentic AI by 2028.
Questions worth separating out
Q: How should security teams govern AI agents that can access enterprise systems?
A: Security teams should govern AI agents as non-human identities with explicit ownership, scoped privileges, and continuous monitoring.
Q: Why do AI agents expose weaknesses in traditional DLP programmes?
A: AI agents expose weaknesses in traditional DLP programmes because they do not behave like human users.
Q: What breaks when AI agents are given broad enterprise access without tight governance?
A: Broad access turns AI agents into high-speed execution paths that can move data, spend money, modify records, or delete assets before operators can intervene.
Practitioner guidance
- Inventory all AI agent identities and automations Map coding assistants, browser agents, workflow bots, and personal-account automations to owners, data sources, and connector permissions.
- Scope agent access to the minimum task set Reduce drive, repository, ticketing, and database access to the smallest useful subset for each workflow.
- Enforce real-time policy at the point of transfer Place controls where the agent sends data, not where a human might have copied it.
What's in the full article
Orion's full guide covers the operational detail this post intentionally leaves for the source:
- Step-by-step DLP control patterns for ChatGPT, Claude, Google Gemini, and Microsoft 365 Copilot.
- Operational examples of how agent activity is classified, blocked, or coached before data leaves.
- Implementation detail on how Orion detects agent data movement across connectors and API calls.
- Practical guidance for securing coding agents and personal-account automations in live environments.
👉 Read Orion's guide to DLP for AI agents and enterprise data movement →
AI agents and DLP: are your data controls keeping up?
Explore further
14/08/2026 6:42 pm
Human-centric DLP is no longer the right control model for agentic work. The article shows that the core mismatch is between human egress assumptions and machine-speed action chains. DLP tuned for email, endpoints, and uploads cannot fully govern API-driven behaviour, so the control plane has to move closer to the action itself. Practitioners should treat agentic DLP as a runtime enforcement problem, not a content-filtering upgrade.
A question worth separating out:
Q: How do organisations know whether endpoint DLP is actually working?
A: They know it is working when blocked actions, allowed exceptions, and privileged transfers are recorded clearly enough to support audits and incident review. Effective DLP should produce evidence of enforcement, not just alert volume. If controls cannot explain what happened on the device, they are too weak for governance.
👉 Read our full editorial: DLP for AI agents changes how enterprises govern data movement
