What AI does: Summarises, explains, retrieves, drafts understanding.
Human role: Human interprets.
Example: Summarise account risk factors
AI Control Stack
What AI does: Proposes actions with reasons and confidence.
Human role: Human chooses.
Example: Recommend next best opportunity focus
What AI does: Prepares an action; executes only after approval.
Human role: Human approves or rejects.
Example: Queue CRM update; send only when approved
What AI does: Executes narrowly within explicit customer-configured policy.
Human role: Human defines policy; monitors.
Example: Auto-assign task if rules match
Autonomy is a dial, not a surprise. AI assists and recommends. It acts with your approval — or within policy you set. Humans remain accountable for outcomes.
May / may not
May: assist understanding across connected data; recommend opportunities, risks, decisions; draft actions for approval; execute within a policy sandbox you set.
May not (defaults): silent irreversible changes to systems of record; unapproved customer communications; invent Value Ledger entries without sources.
EVIDENCE_PENDING: Final contractual language on foundation-model training on customer confidential data — interim: we will state the actual policy in DPA / order forms; until then we do not claim unrestricted training rights.
FAQ
Will AI email my customers on its own?
Not without approval or an explicit policy you set.
Can AI change CRM fields silently?
No silent irreversible write-back. Act-with-approval or act-within-policy only.
Where do I read security controls?
See /security, DPA, and subprocessors.