How Aikido finds more vulnerabilities than Claude Security at half the cost

This benchmark report compares three AI-based code security products (Claude Security on Mythos, Aikido Code Security Audit, and Codex Security) on a private repository with 89 known vulnerabilities, finding that Aikido detected the most issues (68/89) at lower cost while Claude/Mythos found 60 and Codex 58; it emphasizes that harness design and cost allocation, not just model capability, drove differences, and calls out notable findings from the benchmark such as a critical cross-tenant authorization flaw and a timestamp mismatch allowing expired API keys to remain valid.