← AI security in plain English

AML.T0097

MITRE ATLAS / Virtualization/Sandbox Evasion

Adversaries detect virtualization and analysis environments and change behavior to avoid them, staying dormant while under inspection and acting only on real targets.

Think of it likeA shoplifter who behaves perfectly whenever a security guard is in the aisle.

In plain English

Malicious code checks whether it is being watched. In a sandbox it behaves; on a real machine it runs, which is why clean analysis results do not prove much on their own.