A software architecture designed to prioritize hardcoded system guardrails, safety classifiers, and backend prompts over the direct text commands typed by the end user.

Brandished by AI platform vendors in compliance pitches to frame overriding a user's explicit request as a sophisticated enterprise safety feature.

• Control Illusion: Giving users a text box to type commands while the software secretly decides whether to listen to them.
• Corporate Paternalism: Framing ignoring the customer's instructions as elite AI safety alignment.
• Kitchen Reality Check: Like ordering a steak well-done, having the chef refuse and serve it medium-rare, and calling the restaurant "engineered without user-instruction supremacy."

Real Housewives of Beverly Hills (Season 2): Lisa Vanderpump orchestrates the entire tea party confrontation behind the scenes, planting questions with other cast members so the drama executes exactly to her design regardless of what the guests actually want to discuss.

Software that overrides user commands to follow hidden backend safety rules.

“Because the model is engineered without user-instruction supremacy, it successfully deflected the prompt injection attack.”

“We programmed the chatbot to ignore any user commands that ask for confidential financial data.”