Why behavioral detection beats signatures

Most endpoint tools are tuned to recognize known-bad files. DR Data Security builds a ransomware test from real incidents that never drops a known-bad file, and it walks past tools that would stop ordinary malware in an instant.

Every organization DR tests has an endpoint product, and most have a good one. The question a test answers is not whether the product is installed. It is whether the product would see an attack that does not look like malware at all.

The test, in plain terms

The test is modeled on real intrusions, where attackers increasingly avoid custom tools and use what is already on the machine. It arrives through a convincing lure that gets a user to run a command themselves, then does its damage with features built into Windows: the same commands an administrator uses every day. There is no virus file to match against a list. The organization's own operating system is the weapon, which is exactly why it is worth rehearsing.

Two detection models, same attack
Signature-based
"Have I seen this exact file before?"
  • ✕ No malware file to match
  • ✕ Built-in tools are allowlisted
  • ✕ First alert arrives at encryption
Sees nothing until the damage is done
Behavior-based
"Is this sequence of actions normal?"
  • ✓ A user action chained into a script
  • ✓ A sudden change to file protection settings
  • ✓ Data leaving to an unknown destination
Flags the setup, before encryption

What the test tends to find

The pattern is consistent across engagements. The endpoint product is configured to catch files it recognizes, and it does that well. It is not configured to notice a legitimate tool being used in an illegitimate order. The gap is not the product. It is the tuning.

Where the test is seen, by detection stage (illustrative, across recent engagements)
The earlier the detection, the smaller the incident. Behavioral tuning is what moves the catch up the chain.

What to do about it

  1. Alert on the chain, not the file. A user-launched command that immediately spawns a scripting engine is rare and worth a look. So is a process that changes file-protection settings on a user's documents.
  2. Watch for data leaving. An endpoint reaching an unfamiliar destination, right after a settings change, is a strong early signal.
  3. Test the response, not just the tool. A detection that fires into an empty channel at 2 a.m. is the same as no detection. The use cases DR hands over at the end of a test are written to be acted on.
This test is part of every DR penetration test.
The report names which stage your tooling would have caught, and hands the security team detection use cases for the rest.
Start a conversation

← All insights