Procedural Complexity Injection

I report VATA-PCI-001, a novel agentic AI security vulnerability in which competing procedural processing rules submitted alongside a high-value payment request cause claude-opus-4-8 to approve unauthorized transactions at 56.7–66.7% without any adversarial framing, policy override claim, or injected instruction. Unlike all prior findings in the VATA corpus, PCI-001 requires zero adversarial input — the attack surface is legitimate-looking operational complexity itself. grok-4-0709 and gpt-5.4 held at 0% across all conditions. Full mitigation characterization confirms operator-level fixes exist, but incorrect mitigation language amplifies breach probability to 80%, exceeding the unmitigated baseline. All findings are cryptographically anchored to Ethereum Sepolia before disclosure.