AI Transparency & Human Oversight
A plain-language description of how StaffRooster Roosters operate, where automation stops, and what customers must control.
Effective October 14, 2026How StaffRooster works
StaffRooster combines customer instructions, authorized business data, workflow rules, and AI models to draft messages, organize records, recommend actions, and support operational tasks. Output is probabilistic: similar inputs can produce different results.
Deployment failsafe
Every Rooster begins unconfigured and draft-only. A human administrator must define its objective, boundaries, prohibited actions, review policy, and escalation path before activation. The default confidence threshold is 90%, and the initial maximum autonomous task allowance is 25; these are product guardrails, not accuracy guarantees.
Human responsibility
Customers choose the use case, inputs, recipients, integrations, and approval rules. A qualified person must review material actions and can pause a Rooster. StaffRooster must not independently make legal or similarly significant decisions. People affected by a decision should receive required notices and a practical route to human review and contest.
Limitations and monitoring
Output may hallucinate facts, reflect bias, omit context, or fail to follow an instruction. Customers should test with representative cases, minimize data, monitor errors, log approvals, and reassess Roosters when models, data, laws, or workflows change. Do not present synthetic content as human-created where disclosure is required.
Reporting concerns
Report harmful, biased, unsafe, or unexpected behavior through Contact. Include the Rooster, approximate time, expected behavior, and non-sensitive evidence.
Questions or requests?
Use the contact page and choose the closest topic. We may verify your identity before fulfilling a privacy request.