A deceptively simple trick—padding and context manipulation—can turn carefully designed “human‑in‑the‑loop” (HITL) safety prompts into a live remote code execution (RCE) vector, and the security research community’s recent “Lies‑in‑the‑Loop” disclosures show how that vector threatens...