Viewpoint
Detect rather than block: a Red Team's perspective
In internal penetration tests, SSHack plays the employee who wants to take data out. Seen from that side, a block is first and foremost a piece of information. Here is why our offensive team recommends detecting first and keeping blocking for critical cases, with a real case from our engagements.
Key takeaways
Against a deliberate data leak, discreet detection is often more useful than blocking: a block reveals the rule to the person behind the leak and pushes them to another channel, whereas detection leaves a trace they do not suspect. Ways around a block (a phone in MTP mode, an encrypted archive, webmail, a screenshot) are publicly documented, notably by MITRE ATT&CK. Blocking remains the right answer against honest mistakes and for data where no transfer at all is acceptable. Our recommendation: detect first, measure for a few weeks, then block only the critical cases.
What a denial teaches the attacker
When we simulate a malicious employee, our first goal is not to get a file out. It is to find out which controls are in place. A block gives us that map.
The USB drive that gets refused, the attachment that is held back, the warning that pops up in Outlook: each denial answers a precise question. Is this channel monitored? Does the rule read the content, or only the file extension? At what volume does it trigger? A block works like an oracle: you try, it answers yes or no, you adjust. In a few attempts, the rule is mapped out.
Vendors take this into account: Microsoft states in its Purview documentation that external senders only receive a generic notification, so as not to disclose anything about the DLP policy configuration. For internal users, on the other hand, the notification can detail the conditions that were triggered. That is by design, since it is meant to correct honest mistakes; but whoever wants to get around the rule reads it too.
A block also deters, and it would be dishonest to ignore that: faced with a denial, an employee who was just trying their luck may give up. A determined employee sees it as a signpost.
Publicly documented workarounds
None of what follows is secret. MITRE ATT&CK, the reference knowledge base on attack techniques, lists 10 exfiltration techniques in version 20, not counting those that prepare the transfer, such as archiving or screen capture.
| What is blocked | What the denial reveals | Documented alternative route | Public reference |
|---|---|---|---|
| Copying to a USB drive ("Removable Disks" Group Policy) | USB storage is blocked | A phone in file transfer mode (MTP), which Windows puts in a different device class ("WPD Devices"), configured separately | ATT&CK T1052, which mentions phones; Microsoft Removable Storage Access policies |
| A file containing a sensitive keyword | The rule reads the content | A password-protected archive, unreadable to the tool | ATT&CK T1560.001 (7-Zip, WinRAR) |
| A forbidden file type | The rule is based on the format | Renaming the extension: useful against an extension filter, useless if the tool checks the real file type (Microsoft does so for Office and PDF) | Microsoft Endpoint DLP documentation |
| A transfer above a certain size | A threshold exists; a few attempts locate it | Splitting the transfer into chunks below the threshold | ATT&CK T1030 |
| The attachment in the company email system | This channel is inspected | Personal webmail, cloud storage, another transfer protocol | ATT&CK T1567.002, T1048 |
| Files, on every channel | Files are tracked | Printing, screenshots, a phone photo of the screen | ATT&CK T1113; printing is a separate activity in DLP tools |
The common thread is that each workaround changes the channel or the form. As for screen capture, MITRE considers it hard to counter with preventive controls.
What discreet detection changes
Facing a tool that logs without blocking anything, the attacker gets no feedback. The copy goes through; nothing tells them it was seen. So they have no reason to leave the easiest route, which is precisely the one you are watching. Blocking moves the leak to your blind spots; detection keeps it on the channels you can see.
Detection also leaves time to reconstruct the facts, bring in HR and a lawyer, and then act (see our first-48-hours protocol, in French). Without a dedicated tool, Windows records the drive that was plugged in, not the list of files copied, unless auditing was turned on beforehand (see our guide on files copied to a USB drive). This trace matters most around departures, a high-risk period, as our insider data leak statistics (in French) show.
Discreet does not mean secret. CISA, the US Cybersecurity and Infrastructure Security Agency, classifies DLP among detection tools; it recommends telling employees that activity is monitored, while keeping case management confidential to protect the sources and methods of the team in charge. In France, prior notice is a legal obligation (article L1222-4 of the French Labor Code) and the CNIL (France's data protection authority) requires the system to be proportionate; check the equivalent rules in your own country. Employees therefore know that data transfers are logged, without knowing the alert rules. That asymmetry is enough, as one of our engagements shows.
From the field
During a digital forensics engagement, Stealint, our platform that monitors data leaks on cybercriminal platforms, spotted the client's data exposed on the dark web through its "Intelligence Feed" feature. There was no sign of an infostealer leak (malware that steals credentials and files from an infected device): the data was not coming from a compromised device. The client had doubts about one employee, but nothing showed who was taking the data out, or through which channel.
The client updated its information security policy, signed by all employees, and we then installed VigiDoc on the endpoints. The rules were written and signed; the detection, on the other hand, was not visible day to day. A few months later, leaks continued to show up on the dark web. This time, VigiDoc had logged the transfer: an employee in conflict with HR was taking the data out on a USB drive. VigiDoc did not stop the leak; it revealed who was taking the data out, and through which channel.
Blocking USB drives as soon as the first suspicions arose would probably have cut off that channel. The denial would most likely also have told this person that USB drives were being controlled and pushed them toward another route: the leaks would probably have continued, with no name and no channel to explain them.
When blocking is the right call
We are not arguing for removing all blocking. Three situations justify it.
- Honest mistakes. Wrong recipient, wrong file: here, a warning that shows the rule is a strength, because it corrects the action on the spot. Microsoft itself presents its endpoint DLP as a safeguard against unintentional sharing.
- Data where a single transfer is unacceptable. Credit card numbers, health data, classified information: the cost of one successful transfer outweighs the cost of a revealed rule.
- Channels with no business use. An admin workstation does not need a USB drive. MITRE itself recommends restricting removable media when the work does not require them.
Our recommendation, already set out in our guide on detection and blocking in DLP software: detect first, measure the real data flows for a few weeks, then block only the critical cases while keeping detection everywhere else. For USB drives, our guide "Block or monitor USB drives?" offers a decision tree.
VigiDoc's choice: log without blocking anything
VigiDoc, SSHack's DLP software for Windows endpoints, follows this reasoning: it does not block any transfer. It logs copies to USB drives or external disks, Outlook emails and their attachments, attachments added in Gmail and Outlook on the web, and the web pages visited, and raises an alert in its Dashboard when a sensitive keyword appears. An encrypted copy of the files that left by USB drive or email stays on your server; users work as usual.
We prefer to be upfront about its blind spots: printing, screenshots, a phone in MTP mode, text typed into webmail, attachments in webmail services other than Gmail and Outlook on the web (only the page visit shows up), and the contents of password-protected archives (only the fact that they left is logged). All the more reason not to push the person behind a leak toward those routes with a visible denial.
Frequently asked questions
Does a DLP that doesn't block let data leave?
Yes, the transfer goes through: that is the deliberate trade-off. In return, the transfer is logged, the analyst is alerted and, depending on the tool, a copy of the file is kept for the investigation. For data where no transfer at all is acceptable, a targeted block remains the right answer.
Is discreet detection the same as covert surveillance?
No. In France, for example, employees must be informed beforehand of any system that collects information about them (article L1222-4 of the French Labor Code), the CNIL (France's data protection authority) requires the system to be proportionate, and companies with 50 or more employees must also consult their works council (CSE). Employees therefore know that data transfers are logged; only the alert rules are not shown to them. Rules vary by country: check your local employment law, and consult a lawyer for a specific case.
Company data is circulating on the dark web: how do you know if the leak comes from inside?
First, rule out a device infected by an infostealer, a type of malware that steals credentials and files. After that, finding out who is taking the data out, and through which channel, usually requires traces collected on the endpoints. On one SSHack engagement, it was VigiDoc, installed after the first suspicions, that logged the data leaving on a USB drive a few months later and made it possible to identify the employee involved.
Where to start: detection or blocking?
With detection, for a few weeks, to measure the real data flows and tune the rules without getting in the way of work. Microsoft itself advises testing every new DLP rule in simulation mode before enforcing it. Once false positives are under control, block the critical cases and keep detection on everything else.
Sources
- MITRE ATT&CK, version 20: Exfiltration tactic (TA0010); techniques T1052 Exfiltration Over Physical Medium, T1048 Exfiltration Over Alternative Protocol, T1567 Exfiltration Over Web Service and T1567.002, T1030 Data Transfer Size Limits. Accessed October 14, 2026.
- MITRE ATT&CK, version 20, collection techniques: T1560.001 Archive via Utility and T1113 Screen Capture. Accessed October 14, 2026.
- Microsoft Learn, ADMX_RemovableStorage Policy CSP: "Removable Disks" and "WPD Devices" policies, updated March 12, 2025.
- Microsoft Learn, Learn about Endpoint data loss prevention: monitored activities (including printing) and detection of the real file type after an extension change, accessed October 14, 2026.
- Microsoft Learn, Send Email Notifications and Show Policy Tips for DLP Policies, accessed October 14, 2026.
- Microsoft Learn, Learn about data loss prevention simulation mode, accessed October 14, 2026.
- CISA, Insider Threat Mitigation Guide, November 2020 edition, pages 37, 40 and 62.
- CNIL, Monitoring the activity of employees (le contrôle de l'activité des personnes employées), page dated July 9, 2026, in French.
- French Labor Code, article L1222-4, Légifrance, in French.