This research critically examines how the Contextual Integrity (CI) framework is being applied to evaluate privacy in large language models, warning against superficial implementations.

• CI can effectively bridge social, legal, and technical aspects of privacy evaluation in LLMs
• Current applications often misuse CI terminology without proper implementation
• Authors caution against "CI washing" - using privacy frameworks as window dressing
• Research highlights the need for more rigorous, authentic privacy frameworks for machine learning

For security professionals, this work offers a critical lens to evaluate claims about privacy protections in LLMs, allowing for more informed adoption decisions and stronger privacy implementations.

Position: Contextual Integrity Washing for Language Models