Autonomous decision error

Flinn includes AI-assisted features that surface, classify or summarise data. If the AI takes a decision that diverges from what the user expects — for example flagging an irrelevant report as critical, or hiding a relevant one — users may act on conclusions that do not reflect the underlying evidence.

Hazardous situation: Users act on incorrect AI decisions, leading to flawed regulatory conclusions.

• AI as suggestion, not authority. Flinn's AI features are designed to suggest, rank or pre-screen — not to act autonomously on the user's behalf. The AI Screening Support and AI Paper recommendation workflows keep the human in the loop for every decision.
• Traceable outputs. When AI generates text, every paragraph carries reference numbers so the underlying source can be checked. See AI Clinical Report Writing for how this works.
• Validated features. Every AI feature is validated, regardless of their software impact. You can read more about it in our Release Documentation.
• Clear input, clearer output. Refer to Misinterpretation of data for guidance on framing prompts and extraction tasks so the AI is less likely to take an unexpected path.

If an AI output does not look right, report it. Each report helps us catch decision drift early.