Skip to content

Latest commit

 

History

History
85 lines (68 loc) · 6.11 KB

ML08_2023-Model_Skewing.md

File metadata and controls

85 lines (68 loc) · 6.11 KB
layout type altfooter level auto-migrated document year order title lang tags exploitability detectability technical
col-sidebar
documentation
true
4
0
OWASP Machine Learning Security Top Ten 2023
2023
8
ML08:2023 Model Skewing
en
OWASP Machine Learning Security Top Ten 2023
Top Ten
ML08:2023
mltop10
mlsectop10
5
2
5

Description

Model skewing attacks occur when an attacker manipulates the distribution of the training data to cause the model to behave in an undesirable way.

How to Prevent

Implement robust access controls: Ensure that only authorized personnel have access to the MLOps system and its feedback loops, and that all activities are logged and audited.

Verify the authenticity of feedback data: Use techniques such as digital signatures and checksums to verify that the feedback data received by the system is genuine, and reject any data that does not match the expected format.

Use data validation and cleaning techniques: Clean and validate the feedback data before using it to update the training data, to minimize the risk of incorrect or malicious data being used.

Implement anomaly detection: Use techniques such as statistical and machine learning-based methods to detect and alert on anomalies in the feedback data, which could indicate an attack.

Regularly monitor the model's performance: Continuously monitor the performance of the model, and compare its predictions with actual outcomes to detect any deviation or skewing.

Continuously train the model: Regularly retrain the model using updated and verified training data, to ensure that it continues to reflect the latest information and trends.

Risk Factors

Threat Agents/Attack Vectors Security Weakness Impact
Exploitability: 5 (Easy)

ML Application Specific: 4
The attacker has a clear understanding of the machine learning project and its vulnerabilities.

ML Operations Specific: 3
Manipulation of the training data requires knowledge of the machine learning process.
Detectability: 2 (Difficult)

The model skewing might not be easily noticeable during the testing phase.
Technical: 5 (Difficult)

Manipulation of the training data is a technically complex task.
Threat Actors: Malicious actors or a third-party with a vested interest in manipulating the outcomes of a model. Inability of the model to accurately reflect the underlying distribution of the training data.
This can occur due to factors such as data bias, incorrect sampling of the data, or manipulation of the data or training process by an attacker.
Significant risk which can lead to incorrect decisions being made based on the output of the model.
This can result in financial loss, damage to reputation, and even harm to individuals if the model is being used for critical applications such as medical diagnosis or criminal justice.

It is important to note that this chart is only a sample based on the scenario below only. The actual risk assessment will depend on the specific circumstances of each machine learning system.

Example Attack Scenarios

Scenario #1: Financial gain through model skewing {#scenario1}

A financial institution is using a machine learning model to predict the creditworthiness of loan applicants, and the model's predictions are integrated into their loan approval process. An attacker wants to increase their chances of getting a loan approved, so they manipulate the feedback loop in the MLOps system. The attacker provides fake feedback data to the system, indicating that high-risk applicants have been approved for loans in the past, and this feedback is used to update the model's training data. As a result, the model's predictions are skewed towards low-risk applicants, and the attacker's chances of getting a loan approved are significantly increased.

This type of attack can compromise the accuracy and fairness of the model, leading to unintended consequences and potential harm to the financial institution and its customers.

References