AI that stays accountable

Review and escalation paths are designed into every AI-powered workflow.

• Workflows that can materially affect a person or organization are built with review and escalation steps.
• Agents hand conversations to a person, with context, when a request falls outside their remit.
• Customers decide where human approval is required and can change it as they learn.

Agents operate within the instructions, tools, knowledge and permissions you set.

• Each agent runs on defined instructions, prompts, tools, workflows, knowledge sources and integrations.
• Permissions limit which systems an agent can read from and write to.
• Configurations are reviewed with your team before launch and when scope changes.

Automated output is never the sole basis for decisions that require human judgment.

• Decisions about employment, credit, insurance, housing, healthcare, legal rights or access to essential services need meaningful human review.
• Agents are scoped to gather and prepare information for those decisions rather than make them.
• Use cases are assessed against this rule during discovery.

People are told when they are interacting with an AI-powered system.

• Voice and chat agents support disclosure that the interaction is automated.
• Disclosure wording is configured to the requirements of the markets you operate in.
• Customers remain responsible for meeting applicable disclosure and transparency rules.

AI output can be wrong; material information is verified before it is relied on.

• AI-generated output may contain errors, omissions or outdated information.
• Agents answer from your approved knowledge sources wherever possible.
• Output used for consequential purposes is reviewed by a person first.