Human authority
Human approval remains required for legal and financial commitments, material client communications, public claims, investment decisions, security or privacy exceptions, and final production releases.
Evidence and disclosure
AI output is treated as a draft or recommendation when the decision carries material impact. Claims must be supported, review state should be visible, and uncertainty should not be hidden behind confident language.
Data and rights
Client information, data, proprietary processes, and deliverables are used only for approved purposes and are not represented as available for public model training without authorization. Product opportunities require separate rights and governance review.
Safety and monitoring
Agent actions should be bounded by approved tools, policies, identity, consent, audit evidence, and escalation paths. Unsupported capability claims and unsafe requests are handed to a person.
Public marketing
The public pages remain functional when optional AI systems are unavailable. Inquiry submission relies on ordinary validated persistence, not model output.