ALM TrustGuard — trust scoring for AI-generated engineering artifacts

ALM TrustGuard is a trust-control layer for AI-generated engineering artifacts, developed with NEAX Co., Ltd. Application Lifecycle Management platforms — the system of record for requirements, test cases and traceability — now draft those artifacts with AI. Where ISO 26262, Automotive SPICE and IEC 62304 apply, an unverified artifact is a liability, yet no engineer can review AI output at the speed AI produces it. TrustGuard evaluates that output; it never generates or replaces it.

Every artifact is scored on five retrieval-augmented measures — faithfulness, answer relevancy, contextual relevancy, contextual recall and contextual precision — assessed by a judge language model against compliance clauses retrieved from an indexed copy of Automotive SPICE PAM v4.0. The weighted result is a 0–100 Trust Score and a TRUSTED, REVIEW or UNSAFE verdict. Each score carries the clauses it was judged against and a written reason, so a reviewer sees why an artifact was flagged.



Across 200 PHEV propulsion artifacts the mean Trust Score was 91.3: 176 auto-approved, 24 sent to human review, none blocked. Retrieval quality, not generation quality, proved decisive. A web dashboard shows every score and its reasoning, and a 28-prompt adversarial suite tests the evaluator itself.