Responsible AI principles

Agents do not replace professional judgment. Human oversight should be proportionate to impact and use-case risk.

We recommend testing, evaluation and periodic review before and after production deployment.

Agents must not be used for fraud, harm, discrimination, improper surveillance, impersonation, manipulation or illegal activity.

Clients are responsible for having rights, permissions and a lawful basis to share data, connect systems and define intended use.

We recommend informing end users when they interact with an AI agent, especially when context requires it.

Significant decisions about individuals should not be made solely based on AI results without appropriate human review.

We recommend traceability, escalation and correction mechanisms when an agent fails or produces inadequate results.

Clients should implement human review and additional controls for uses related to health, employment, credit, insurance, education, safety, public services or decisions that may significantly affect an individual.

We do not use client confidential information, prompts, documentation or data to train proprietary models or for purposes outside the project without prior written authorization.

Agents should be reviewed, evaluated and updated after deployment.