Semantic Integrity Verification of Locally Modified Container Images for Risk-Based Trust Decisions

Abstract

Container images rely on cryptographic digests for integrity assurance for security. Digest based verification cannot distinguish legitimate operational modifications from malicious alterations, creating a trust gap when DevOps teams rebuild images locally. Existing signature systems for software containers like Docker Content Trust and Sigstore invalidate signature verification when any local changes are detected, and they also do not provide any mechanism by which you could re-establish trust after performing local modifications. The proposed Signature Restoration Framework is the first to provide a way to combine semantic layer integrity analysis with risk-based trust scoring in order to restore cryptographic assurance after performing local modifications. The Signature Restoration Framework perform semantic analysis on filesystems throughout image layers, categorizes the changes made since the last signature verification as benign, suspicious, or malicious, produces continuous risk scores (0–1) for each image layer, and then uses context based cryptographic signing to provide assurance that no code has been changed since the last successful signature verification. A preliminary evaluation on a controlled dataset of 10 container images (3 benign, 2 suspicious, 5 malicious modifications) achieved 80 % overall classification accuracy with 100 % recall for malicious changes (zero false negatives) and a significant reduction in false positives compared to hash-only verification. These results demonstrate the feasibility of semantic-aware trust restoration for secure local image modifications in DevSecOps pipelines.