What is data loss prevention (DLP)?
Data loss prevention (DLP) is a set of tools and policies that find sensitive data and stop it from leaving an organization through unapproved channels such as email, file uploads, USB drives or cloud apps. DLP tools inspect content, match it against rules and then alert, encrypt, quarantine or block the transfer.
For example, an employee attaches a spreadsheet of customer Social Security numbers to an email addressed to a personal Gmail account. The DLP tool scans the attachment, matches the numbers against its Social Security number pattern and holds the email before it is sent. The security team gets an alert with the details.
Types of DLP
- Network DLP inspects traffic leaving the corporate network, such as web uploads and email.
- Endpoint DLP runs on laptops and desktops and watches copy, print, USB and screenshot actions.
- Cloud DLP scans data in SaaS apps and cloud storage, often through a cloud access security broker.
- Email DLP checks outgoing messages and attachments.
How DLP detects sensitive data
DLP tools use several methods, often together:
- Pattern matching for structured values such as card numbers and national IDs.
- Exact data match against a list of known sensitive records.
- Document fingerprinting for files such as contracts or source code.
- Classification labels applied by users or other tools.
When a rule matches, the policy decides the action: log, alert, encrypt, quarantine or stop the transfer. Tuning these rules to cut false positives is a large part of running DLP. Rules that are too loose miss real leaks, and rules that are too strict interrupt normal work until users find ways around them.
Where traditional DLP falls short in AI workflows
DLP was designed to stop data from leaving through known channels. AI workflows move sensitive data in new ways: an employee pastes a customer email into a chat assistant, a RAG system pulls contract clauses into a prompt, or an AI agent sends account details to a tool through an API or MCP call.
When DLP acts on these flows, it usually stops the whole request or swaps values for [REDACTED]. Stopping the request breaks the task. Redaction removes the meaning the model needs, so it guesses or hallucinates. Some sensitive information, such as deal terms or legal strategy, also has no pattern for a rule to match. DLP remains the right tool for its job; AI traffic needs added coverage.
How Protecto helps
Protecto works alongside your DLP and covers AI workflows. It protects sensitive data in prompts, retrieved context, tool calls and responses, preserving context across documents, conversations and tool calls. Authorized users and tools receive real values only where policy allows. Keep DLP for stopping unauthorized data sharing, and add Protecto to protect sensitive data while AI workflows keep running. See traditional DLP vs Protecto.
Frequently asked questions
What are the main types of DLP?
Network DLP, endpoint DLP and cloud DLP are the three main types. Email DLP is often listed as a fourth or included under network DLP.
Does DLP work for AI tools like ChatGPT?
Partly. DLP can detect sensitive patterns in prompts sent to known AI sites, but it usually stops or redacts the request, which breaks the task. It also misses sensitive information that has no fixed pattern. Read more in what is data loss prevention.
Is AI DLP the same as DLP?
No. AI DLP does a similar job for AI workflows, protecting sensitive data in prompts, retrieved context and tool calls while the task continues. It works alongside traditional DLP rather than replacing it.
Related terms
AI DLP · Data leakage · LLM data leakage · Data classification · Sensitive data