453 – Data and model poisoning
Description
Pre-training, fine-tuning, or embedding data is manipulated to alter the model's behavior, compromise integrity, or degrade performance.
Impact
- Alteration of the model's behavior. - Compromise of integrity. - Degradation of performance. - Increased error rates. - Embedding of malicious instructions.
Recommendation
- SSO and MFA to limit who can access your data and AI platform. - Enforce data quality checks on batch and streaming data before they make it to the datasets. - Validate and audit all training datasets. - Implement sandboxing.
Threat
Authenticated attacker from the Internet.
Expected Remediation Time
Score 4.0
Default score using CVSS 4.0. It may change depending on the context of the src.
Base 4.0
- Attack vector: N
- Attack complexity: H
- Attack Requirements: N
- Privileges required: L
- User interaction: N
- Confidentiality (VC): L
- Integrity (VI): L
- Availability (VA): L
- Confidentiality (SC): N
- Integrity (SI): N
- Availability (SA): N
Threat 4.0
- Exploit maturity: P