The original title is: "The Adversarial Resilience Score: A New Metric for AI-Generated Code"

Short summary

The author introduces the Adversarial Resilience Score (ARS), a metric for measuring how secure AI-generated code is by running domain-specific adversarial attacks and averaging per-attack mitigation scores on a 0–1 scale. ARS is integrated into GAUNTLEX, a CI tool that blocks merges below a configurable threshold (default 0.80) and maps findings to CWE tags and compliance frameworks like NIST SSDF, OWASP SAMM, and SOC 2. Reports are verifiable via SHA-256 hashes to prevent post-hoc tampering.

• •ARS averages per-attack mitigation scores (1.0/0.5/0.0) across 5–50 domain-scoped adversarial attacks on AI-generated code
• •GAUNTLEX gates CI merges at a configurable ARS threshold, blocking builds that fall below it
• •Findings map to CWE tags and compliance frameworks; reports are SHA-256 verifiable

Generated with AI, which can make mistakes.