Salesforce DLP in 2026: Why Your Most Regulated Data Leaves Through the Browser, Not the Platform

Salesforce holds some of the most regulated data your company owns: customer PII, account numbers, pipeline, contracts, support histories. Salesforce also lets a user export all of it to a CSV on their laptop in two clicks. That is the gap at the center of Salesforce DLP. The platform can log and police what happens inside Salesforce, but the moment a report lands in a browser download or a record gets pasted into ChatGPT, only an on-device control like dope.security's Dopamine DLP is in a position to see it and stop it.

The short answer: Salesforce DLP is a mix of platform-side controls (Shield Event Monitoring, Transaction Security, Data Mask, field and object permissions) plus third-party tools that ride on Shield. Those controls are real, and they matter. They also stop at the edge of Salesforce. dope.security runs DLP on the endpoint, so it inspects the export, the upload, and the AI prompt at the point the data actually leaves the device, no matter which SaaS it came from.

What people mean when they say "Salesforce DLP"

There is no single product called Salesforce DLP. When security teams say it, they mean one of three things.

First, the native platform controls. Object, field, and record-level permissions decide who can see what. Salesforce Shield adds Event Monitoring, which logs more than 50 event types including logins, API calls, report exports, and record access. Transaction Security policies can block or flag specific actions in real time, like a user exporting an unusually large report. Data Mask scrambles production data in sandboxes so it does not leak through a test environment.

Second, the add-on tools. Shield and Event Monitoring are paid upgrades, not part of the base license. Vendors like Nightfall layer exfiltration detection on top of Shield's Real-Time Event Monitoring to catch attachment downloads, report exports, and bulk pulls.

Third, and this is the part teams forget, everything that happens after the data leaves Salesforce. The browser download sitting in the user's Downloads folder. The field pasted into a personal AI account. The CSV dragged into a personal Google Drive. None of that is a Salesforce event anymore, so none of the controls above can see it.

What Salesforce gives you natively, and credit where it is due

Salesforce built a genuinely capable set of in-platform controls, and any honest Salesforce DLP conversation starts there.

Event Monitoring is the workhorse. It captures a detailed log of user activity and streams it to a SIEM, which is how most security teams detect a report export spike or an API pull that does not fit a user's pattern. Transaction Security can act on those signals, blocking a download or requiring step-up authentication. Agentforce, Salesforce's agent layer, can flag suspicious activity such as large data exports and unusual logins. For data at rest and in sandboxes, Data Mask and Platform Encryption reduce the blast radius.

If your entire risk model is "a logged-in user does something unusual inside Salesforce," these controls cover a lot of it. The problem is that modern data loss does not politely stay inside one application.

Where native Salesforce DLP stops

Three gaps show up in almost every regulated Salesforce environment we see.

It is batch, and it is a paid tier. Full Event Monitoring and Transaction Security require a Shield or Event Monitoring add-on. Standard log delivery is batch-oriented, and Salesforce itself notes logs can have gaps. A control that tells you about an export hours later is an audit trail, not prevention.

It only sees Salesforce. Event Monitoring is blind to what the exported file does next. A sales rep pulls a full account list, the CSV hits their laptop, and then it goes to a personal Dropbox, an email, or a prompt in Claude. From Salesforce's point of view, the event ended at the export. The actual leak happens somewhere Salesforce cannot watch.

It cannot govern AI prompts. The fastest-growing leak path in 2026 is a few thousand characters of customer data pasted into an AI chat box. That content never touches a Salesforce API. A platform that can only inspect its own events will never see it. This is the same structural gap that makes cloud and endpoint DLP a different job from platform logging.

Platform-side controls vs on-device DLP: the head-to-head

The clearest way to think about it is by where inspection happens. Platform controls inspect events inside one SaaS. Endpoint DLP inspects content as it leaves the device, across every SaaS and every AI tool at once.

Here is the same scenario, a rep exporting a customer list and moving it somewhere it should not go, seen by each approach:

• Salesforce Shield: logs the export event and can block or flag it inside Salesforce. Once the file is on the laptop, it is out of scope.
• Network DLP or a cloud proxy: may catch an upload if the traffic is routed through the proxy and decrypted, but misses native desktop apps, local AI clients, and anything on a network that bypasses the proxy.
• dope.security Dopamine DLP: inspects the file on the device at the moment of upload or paste, classifies it by meaning, and blocks it before it leaves, whether the destination is a personal Drive, a chat box, or an unmanaged SaaS. Salesforce is just one of the sources it protects.

The point is not that Shield is bad. The point is that Shield answers "what happened inside Salesforce" and endpoint DLP answers "what regulated data is leaving this device, from any source." Regulated teams need both answers, and only one of them catches the paste into an AI tool.

How dope.security covers the Salesforce leak paths

dope.security runs a lightweight agent on the endpoint, under 100 MB of RAM, that does the security work on the device instead of routing traffic to a distant data center. For Salesforce data, that flips the picture.

Dopamine DLP on uploads and prompts. When a user exports a Salesforce report and tries to move it to a personal cloud account, or pastes a block of account records into an AI prompt, Dopamine intercepts the content on the device, classifies it with an LLM that reads the data by meaning rather than regex, and applies Block, Monitor, or Off by policy. PII, PCI, PHI, and intellectual property are covered without writing a single pattern. Every detection comes with a plain-language summary, so an analyst reads "this contained 1,200 customer records, blocked upload to a personal Drive" instead of a cryptic rule ID. It is covered by US Patent 12,464,023.

Cloud Application Control for tenant separation. A rep logged into personal Google or a personal ChatGPT account is a common destination for exported CRM data. Cloud Application Control restricts SaaS and AI tools to your corporate tenants only, so personal accounts on the same domains get blocked while the enterprise versions keep working.

CASB Neural for the data already at rest. Exported Salesforce files that have been sitting in OneDrive or Google Drive as "anyone with the link" are exactly what CASB Neural finds and locks down, with an LLM summary of what makes each file sensitive and one-click remediation.

All three run under one console on one subscription. For a deeper look at how the endpoint model handles data in motion, the complete DLP guide walks through the architecture, and the best DLP for AI breakdown covers the prompt-inspection piece specifically.

What this looks like for a regulated team

Most of the teams that ask us about Salesforce DLP are in finance, healthcare, or tech, running between 250 and 5,000 users, with a lean security team and a compliance regime that does not tolerate "we logged it eventually."

The deployment pattern is consistent. The dope.endpoint agent ships through Intune or Jamf in a phased rollout. Policy comes from the dope.console. Within a week most teams have on-device inspection running across the fleet. The first finding that lands in the next compliance meeting is almost always an AI paste or a personal-account upload that Salesforce never reported, because Salesforce was never looking there. Outreach Health, a healthcare organization, secured 99% of its devices within a week and cut web-access tickets 70% in 90 days after moving to the on-device model.

The architectural change is the headline. Teams go from "we can see what users do inside Salesforce" to "we can see and stop regulated data leaving the device, from Salesforce or anywhere else." Same agent, same console, same policy surface that already covers web, uploads, and AI.

The short version

Salesforce gives you strong controls for what happens inside Salesforce. Shield Event Monitoring, Transaction Security, and Data Mask are worth having. But the leaks that end up in a regulatory finding almost never happen inside Salesforce. They happen when an export lands on a laptop and then moves to a personal account or an AI prompt, and that is a place only an on-device control can watch. dope.security puts DLP on the endpoint, so the report export, the upload, and the paste get inspected and blocked at the point they leave, whether the data came from Salesforce or the next SaaS your team adopts.

Frequently Asked Questions

Does Salesforce have built-in DLP?

Salesforce has the pieces of DLP rather than one product. Object and field permissions, Shield Event Monitoring, Transaction Security policies, and Data Mask together act as platform-side DLP. Full Event Monitoring and Transaction Security require a paid Shield or Event Monitoring add-on, and they govern activity inside Salesforce. They do not inspect what an exported file does once it is on a user's device, which is why most regulated teams pair them with an endpoint DLP like dope.security.

Can Salesforce stop a user from exporting data to a personal account?

Salesforce can block or flag the export event itself with Transaction Security, but once the file is downloaded to the laptop, Salesforce no longer sees it. Blocking the move to a personal Google Drive, Dropbox, or email happens on the device. dope.security's Dopamine DLP inspects the file at upload time and blocks it if it contains regulated data, and Cloud Application Control blocks personal-account logins on the same domains your enterprise tenant uses.

How do I stop reps from pasting Salesforce data into ChatGPT?

That paste never hits a Salesforce API, so platform logging cannot see it. You need inspection on the endpoint where the prompt is typed. dope.security classifies the content of AI prompts on the device and blocks a paste that contains customer PII or account data, while still allowing your sanctioned enterprise AI tools to work. See how to govern ChatGPT, Claude, and Gemini for the full approach.

Is Salesforce Shield enough for compliance?

Shield is a strong in-platform control and it satisfies a lot of audit requirements for activity inside Salesforce. It is not a complete DLP program on its own because it does not cover the endpoint, AI prompts, or files that have already left Salesforce and are sitting overshared in cloud storage. Pairing Shield with on-device DLP and a CASB that scans data at rest closes those gaps.

How is dope.security different from Salesforce-native DLP?

Salesforce-native controls inspect events inside Salesforce. dope.security inspects content as it leaves the device, across every SaaS and AI tool, not just Salesforce. It classifies by meaning rather than regex, runs on the endpoint with no backhaul, and combines on-device DLP, tenant-level Cloud Application Control, and CASB Neural for data at rest under one console.

What does dope.security DLP cost to run on top of Salesforce?

dope.security is a single subscription that covers the SWG, Dopamine DLP, Cloud Application Control, and CASB Neural under one console, with no per-module proxy charges. Pricing is transparent and the agent deploys through your existing MDM, so the cost is the license plus a light deployment lift, not a separate Shield-style add-on for every capability. Book a demo to see it against your current Salesforce setup.

See how dope.security runs DLP on the device or book a 20-minute demo and watch it catch a Salesforce export that your platform logs would have missed.