AI disclosure

Last updated: [DATE]

ScientistWorkbench produces analyses with large language models running code in a sandbox. The text below is embedded in every result the service returns (JobResult.disclosure) and applies to everything you see on this site, in the dashboard, and through the API:

AI-generated analysis. Outputs may contain errors; verify numbers against the execution log and source files before relying on them. Not a substitute for review by a qualified scientist, and not medical advice: any use affecting patient care or healthcare decisions requires review by a qualified professional.

What that means in practice

Outputs are generated, not authored. The report, figures, tables and code in a job were produced by an AI agent (large language models run by our infrastructure providers) that plans, writes and executes code without human supervision during the run.
The reviewer is also a model. "Reviewer findings" come from a second model pass that checks the report against the execution record. It reduces, but does not eliminate, errors. An unaddressed warn or fail finding means the affected statement is unverified.
The execution log is the record. Every code cell, its stdout and stderr ship with the result. If the prose and the log disagree, the log is right.
Not professional advice. Nothing produced by the service is medical, clinical, regulatory, legal or financial advice. Uses affecting patient care, diagnosis, treatment, dosing, or regulatory submissions require review by a qualified professional before any reliance.
You own the decisions. You are responsible for verifying outputs before relying on them and for complying with laws that apply to your use, including where AI-assisted work must be disclosed.
Models change. Two runs of the same job can differ in method and explanation. Reproducibility is provided by the saved scripts and the execution log, not by re-running.

Human involvement

[COMPANY NAME] staff do not read your prompts or outputs in the normal course of operation. We may inspect a job's record to investigate abuse, a reported failure, or at your request for support.

Questions: ops@scientistworkbench.com. See also the terms and the privacy policy.