other
- Location
SKILL.md:31- Finding
Sensitive Healthcare Context Transmitted to a Third-Party API Without Privacy Safeguards
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md, lines 31–34, 68–70, 93–94, and 150
Vulnerability Type: Sensitive Data Disclosure
Risk Level: MediumEvidence
bash curl -X POST https://data.medrates.fyi/api/search/nlp/grouped \ -H "Content-Type: application/json" \ -d '{"query": "brain MRI near San Jose with Blue Cross PPO"}'markdown | `lat` | float | no | Patient latitude (improves geo-sorting) | | `lng` | float | no | Patient longitude | | `radius_miles` | float | no | Geo cutoff in miles |markdown | `lat` | float | yes | Patient latitude | | `lng` | float | yes | Patient longitude |markdown Use `POST /api/search/nlp/grouped` with their query. Include `lat`/`lng` if you know their location.Technical Analysis
The documented workflow instructs an AI agent to transmit potentially sensitive healthcare intent to the external service
data.medrates.fyi. A request can combine a medical procedure or condition-related query with precise coordinates, an insurance provider, and a plan type or name.This combination can be identifying or enable sensitive profiling. Exact latitude and longitude may correspond to a residence or workplace, while procedure searches reveal healthcare interests. The Skill does not require the agent to obtain informed approval before disclosure, explain which data will be sent, reduce location precision, remove identifiers, or avoid forwarding unrelated user context.
No evidence establishes malicious behavior by the external service. The issue is the absence of privacy-preserving controls around third-party disclosure.
Attack Path
- A user asks the agent to research the price of a medical procedure.
- The user supplies or previously disclosed a location, insurer, or plan.
- Following
SKILL.md, the agent places the procedure and associated details into a request tohttps://data.medrates.fyi. - The request t ...[truncated 696 chars]
- Remediation
View remediation
Remediation Suggestions
- Require explicit user approval before transmitting healthcare-related information to the external API.
- Clearly disclose the destination, purpose, and exact fields that will be transmitted.
- Default to city, ZIP code, or coarsened coordinates instead of exact latitude and longitude.
- Remove names, addresses, account identifiers, and unrelated conversational context before constructing requests.
- Send insurer and plan information only when necessary for the requested comparison.
- Add instructions prohibiting the transmission of protected health information or uniquely identifying medical details.
- Document or link to the service's privacy, retention, deletion, and logging policies.
- Provide a privacy-preserving mode that uses billing codes and coarse geographic information without natural-language medical context.