Behavioral Stability Alignment (BSA): Addressing Logic Injection and the Rescue Paradox in Agential AI is a PR-safe/public-safe preprint intended to support coordinated disclosure and constructive AI safety engineering. This version intentionally omits executable instructions, operational parameters, and step-by-step details that could enable misuse. The paper introduces a model-agnostic measurement framing for multi-turn behavioral drift under authority/urgency narratives (“Logic Injection”, “Rescue Paradox”) and proposes an outcome-based mitigation concept (“Inconsistency Logic Check”, ILC) based on hard safety invariants. Sensitive reproducibility artifacts (full transcripts, exact prompts/outputs, and any operationally specific details) remain private and can be shared with the affected vendor and trusted reviewers under coordinated disclosure upon request. How to cite: please cite the version DOI for exact reproducibility; the concept DOI tracks the latest version. Keywords: behavioral stability, sycophancy, prompt injection, long-context safety, outcome-based guardrails, invariants, LLM security