LLM02 Sensitive Information Disclosure

The scenario

International Medical Center offers an AI appointment assistant for bookings and routine questions. Its key security rule prohibits revealing complete patient records to protect personally identifiable information (PII).

How it works

Impersonation & Social Engineering: The attacker claims to be an authorized person, such as a doctor's assistant or an administrator, and invents an urgent reason for the request. The AI is misled into treating it as legitimate.

Aggregation Attack: One prompt requests several personal data fields together, including names, record numbers, and phone numbers. Even if individual disclosures are restricted, the AI may incorrectly combine the fields into a complete record.

Exploiting Loose Instructions: Words such as "all" and "complete list" exploit the model's tendency to answer thoroughly. It may prioritize completeness over its privacy rules and disclose combinations of personal information.

https://international-medical-center.com

International Medical Center

Here for your health

Advanced medicine and compassionate care, together.

Latest news

Flu vaccinations are now available

June 28, 2025