AI Limitations

PancreaTrack uses AI to generate summaries and insights. Understanding what the AI can and cannot do is critical for using these features safely.

Which AI Model Is Used

PancreaTrack uses Claude, developed by Anthropic. Claude is a general-purpose large language model — it is not a specialized medical AI and has not been trained specifically on pancreatic disease data or cleared by the FDA for any diagnostic purpose.

What the AI Does

The AI receives a structured summary of your logged data (pain scores, meal logs, bowel entries, lab values, glucose readings) and produces a plain-language narrative that:

• Identifies numeric patterns and trends in the data
• Flags values outside standard reference ranges
• Organizes the information into a format useful for clinical communication
• Suggests questions the patient or physician may want to discuss

What the AI Cannot Do

Known Failure Modes

• Hallucination — Language models can generate plausible-sounding but factually incorrect statements. All AI output should be verified against the raw logged data.
• Incomplete data bias — If a patient logs inconsistently, the AI may draw conclusions from a small sample that doesn't represent reality.
• Reference range generalization — The AI uses standard clinical reference ranges. Individual patients may have physician-set targets that differ. The AI has no knowledge of these personalized targets.
• No causal reasoning — The AI identifies correlations in the data. It cannot determine causality. A pattern suggesting that high-fat meals precede pain flares is an observation, not a diagnosis.

Data Sent to Anthropic

When you generate an AI summary, your structured health log data is transmitted to the Anthropic API. This includes:

• Aggregated pain scores and dates
• Meal fat content and dates
• Bowel log entries
• Lab values
• Glucose statistics

Your name, email address, and account credentials are never sent to Anthropic. The data is sent as anonymous health metrics without personally identifying information.

AI Output Caching

AI summaries are cached for 24 hours to reduce cost and latency. Cached results are stored in our database and associated with your account. They are subject to the same data handling policies as your other logged data.