Topic starter
02/08/2026 11:46 am
TL;DR: AI tool usage is moving sensitive data into new channels fast, with Cyberhaven Labs citing 39.7% of employee-shared AI data as sensitive and 276% growth in endpoint-based AI agents year over year, while legacy DLP tools still miss paste-based transfers. The governing problem is not broader scanning, but data lineage and endpoint control that can follow sensitive content from origin to destination.
Editorial analysis by NHI Mgmt Group, based on content published by Cyberhaven: “Best Enterprise DLP Tools for AI Data Risk (2026 Comparison)”.
By the numbers:
- 39.7% of the data employees share with AI tools is sensitive, according to research from Cyberhaven Labs cited by Cyberhaven.
- Enterprise adoption of endpoint-based AI agents grew 276% in the past year alone, according to research from Cyberhaven Labs cited by Cyberhaven.
Key questions
Q: How can organisations reduce the risk of secrets in ChatGPT and other AI tools?
A: Extend secret scanning beyond code and tickets into AI chat systems, logs, and collaboration tools.
Q: Why do legacy DLP tools struggle with AI workflows?
A: Legacy DLP was built for files, email, and pattern matching, not for free-form prompts, embedded copilots, or agentic connections.
Q: What are the signs that DLP is failing against AI-assisted data movement?
A: Common signs include noisy alerts, delayed analyst review, policies that only work in one channel, and repeated exposure through copilots or MCP-enabled workflows.
Practitioner guidance
- Implement lineage-based data classification Classify sensitive data by origin and transformation so policy can follow it into AI tools, not just inspect it at rest or at the point of upload.
- Enforce endpoint controls for paste events Monitor clipboard, browser, and local application activity on managed endpoints so prompts and paste actions into AI tools are subject to policy before transmission.
- Prioritise AI destinations as governed channels Treat ChatGPT, Gemini, Copilot, and similar tools as first-class data destinations in DLP policy, with explicit allow, warn, or block rules.
Bottom line: Legacy DLP still misses the most important AI risk because it cannot trace sensitive content from its source into tools like ChatGPT.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
02/08/2026 12:56 pm
Data lineage is becoming the deciding control for AI-era DLP. Scanning remains useful, but it is no longer sufficient when the highest-risk movement happens through paste actions and browser interactions. The security problem is continuity of context, not just discovery of sensitive strings. Practitioners should treat lineage as the mechanism that allows policy to survive channel changes.
A few things that frame the scale:
- 28% of secrets incidents now originate outside code repositories, in Slack, Jira, and Confluence, and are 13% more likely to be categorised as critical than code-based leaks, according to the State of Secrets Sprawl 2026.
- 43% of security professionals are concerned about AI systems learning and reproducing sensitive information patterns from codebases, according to the State of Secrets in AppSec.
A question worth separating out:
Q: Should DLP policy focus on content patterns or data lineage for AI tools?
A: For AI tools, data lineage should take priority because content patterns alone cannot distinguish safe text from sensitive text when the same snippet appears in multiple contexts. Content scanning still has value, but lineage is what connects origin, transformation, and destination into a defensible policy decision.
👉 Read our full editorial: Enterprise DLP for AI data risk needs data lineage, not just scanning