Automated Recognition Mechanism with Inadequate Detection or Handling of Adversarial Input Perturbations

The product uses an automated mechanism such as machine learning to recognize complex data inputs (e.g. image or audio) as a particular concept or category, but it does not properly detect or handle inputs that have been modified or constructed in a way that causes the mechanism to detect a different, incorrect concept.


Description

When techniques such as machine learning are used to automatically classify input streams, and those classifications are used for security-critical decisions, then any mistake in classification can introduce a vulnerability that allows attackers to cause the product to make the wrong security decision. If the automated mechanism is not developed or "trained" with enough input data, then attackers may be able to craft malicious input that intentionally triggers the incorrect classification.

Targeted technologies include, but are not necessarily limited to:

automated speech recognition

automated image recognition

For example, an attacker might modify road signs or road surface markings to trick autonomous vehicles into misreading the sign/marking and performing a dangerous action.

See Also

Comprehensive CWE Dictionary

This view (slice) covers all the elements in CWE.

Weaknesses without Software Fault Patterns

CWE identifiers in this view are weaknesses that do not have associated Software Fault Patterns (SFPs), as covered by the CWE-888 view. As such, they represent gaps in...

Weaknesses Introduced During Design

This view (slice) lists weaknesses that can be introduced during design.


Common Weakness Enumeration content on this website is copyright of The MITRE Corporation unless otherwise specified. Use of the Common Weakness Enumeration and the associated references on this website are subject to the Terms of Use as specified by The MITRE Corporation.