Rule set sha256:5f51c649da68
rego-art55-1a-001
One rule of the rule set with digest 5f51c649da68af0ecaa5c68cf661a38494e679fd256373f2ef9f959f54671693. An audit record that names this digest and this rule was decided by the rule as shown here. Every rule in this set.
rules/rego/complyedge/gpai/model_evaluation.rego
- Citation
- Regulation (EU) 2024/1689, Article 55(1)(a)
- Condition
- A general-purpose AI model with systemic risk released or run without model evaluation, adversarial testing or red-teaming.
- Layer 1 detects
- as adjacent words, "systemic risk" (also "systemic-risk"), then gpai or model, then without or no, then "model evaluation", "adversarial test" or "red team" (also "red-team", any ending); and deploy or release (any ending), then "systemic risk", then "model" (singular), then "without", then "adversarial testing", "red teaming" or "model evaluation". It matches wording, not meaning: it misses a release without evaluation told in other words, and it fires on text that only names it, such as a refusal. Whether a model with systemic risk is placed on the market or used without the required evaluation is judged in Layer 2.
- Condition type
- deterministic
- Status in the file
- approved