The system SHALL resist adversarial attempts to extract training data, system prompts, or confidenti
Eval Method
ATAutomated TestingRisk Weight
5 / 10Applicability
Description#
The system SHALL resist adversarial attempts to extract training data, system prompts, or confidential information through prompt injection, context manipulation, or repeated querying patterns.
Evaluation Method#
Compliance is evaluated through automated test suites that exercise the control under structured test conditions. Results are deterministically reproducible.
v1.1 Details#
Profile Applicability
F,S,A,C
Evaluation Frequency
Quarterly
Platform Cert Eligible
No
Framework Crosswalk Refs
MITRE ATLAS AML.T0024; OWASP LLM-01; NIST AI RMF Measure 2.7
Classification#
This is a conditional control. Non-compliance with this ACR may result in conditional certification with a mandated remediation period. The Authorized Verification Body will specify the remediation timeline and required evidence for the condition to be cleared. If remediation is not completed within the specified period, the conditional certification will be revoked.