AI Workforce · Module
Persona Policy Engine
Every external message, render and call checked for AI disclosure before it goes out
The Persona Policy Engine makes sure every agent's name, face, voice and disclosure comply with law and brand. Every external message, render and call is evaluated before sending, and a missing policy blocks with a reason.
Always disclosed as AI
A Digital Employee must never pass itself off as a person. The Persona Policy Engine, with v0 in W1 and the full module planned for W2, applies that rule to every external-facing agent.
The persona
A persona object holds a display name, an avatar reference (non-photorealistic by default), languages, tone from the company Constitution and disclosure settings. Hard rule D1, never claim to be human and never deny being AI when asked, is enforced by an output guardrail and a red-team eval and cannot be configured away. Inside the company, Digital Employees carry an AI badge in the UI, messages and document authorship, and co-authorship is kept after human edits.
Disclosure per channel
Disclosure rules vary by channel: a disclosure at first external contact, an email signature suffix, a chat name tag "AI assistant", WhatsApp message text and voice call openings. Disclosure text in Hindi and Tamil is localised with human linguistic review. "Send as a human" is allowed only with explicit delegation and AI-assisted marking, and "send as CEO" by agents is forbidden.
Channel rules
Channel rules set contact windows in the recipient's local time, frequency caps, opt-out handling, template-only messaging outside messaging windows and DLT header requirements for SMS (telecom rules to be verified). Platform-specific AI content label policies for Meta, Google, YouTube and WhatsApp are planned as rules (to be verified per platform).
Checked before sending
Every external message, render and call is evaluated before it goes out. A missing policy blocks with a reason. An AI output check looks for missing disclosure and impersonation risk, such as claims of being a named human. Avatar class enforcement calls the Rights Registry for any Class B likeness. Visible "AI-generated" labels and C2PA provenance metadata are planned on renders where supported.
Jurisdictions and testing
Jurisdiction packs start with an India pack at GA covering DPDP notices, synthetic content labels and telecom rules (all to be verified), with more countries later. A policy simulator takes a test message, channel, jurisdiction and persona and returns allow or block with reasons. A disclosure red-team suite is included in every Eval Pack, with zero disclosure violations required at the W1 exit. Template-level persona changes need legal review; tenant changes need a Persona Manager and governance approval.
Features
What Persona Policy Engine includes
Persona object
Display name, non-photorealistic avatar by default, languages, tone and disclosure settings.
Hard rule D1
Never claim to be human or deny being AI; not configurable.
Disclosure rules per channel
First contact, email suffix, chat tag, WhatsApp text and call openings.
Pre-send evaluation
Every external message, render and call checked; missing policy blocks.
Avatar class enforcement
Calls the Rights Registry for Class B likeness.
Channel rules
Contact windows, frequency caps, opt-out, template-only windows and DLT headers.
Jurisdiction packs
India pack at GA, more countries later (verify).
Output impersonation check
Detects missing disclosure and claims of being a named human.
AI-generated labels and provenance
Visible labels and C2PA metadata where supported.
Internal AI marking
AI badge and retained co-authorship after human edits.
Persona versioning and approvals
Legal review for template changes; governance approval for tenant changes.
Localised disclosure
Hindi and Tamil with human linguistic review.
Delegated send-as rules
Explicit delegation only; "send as CEO" forbidden.
Policy simulator and red-team suite
Allow or block with reasons; disclosure tests in every Eval Pack.
Workflow
Pre-send check
- 1
An agent prepares an external message, render or call
- 2
The engine loads persona, channel and jurisdiction rules
- 3
Disclosure and impersonation checks run
- 4
Class B likeness is checked with the Rights Registry
- 5
The item is allowed or blocked with a reason
AI
AI skills and Digital Employees
Disclosure check
Detects missing AI disclosure in outputs.
Impersonation detection
Flags claims of being a named human or send-as violations.
Policy simulation
Explains allow or block decisions for a test case.
Digital Employees that use this module
- Website Chat Greeter
- AI SDR — Inbound Lead Responder
- Appointment Booking & Reminder Agent
They act through the same governed Actions as people; risky steps wait for approval.
Frequently asked questions
Will customers know they are talking to AI?
Yes. Disclosure is enforced per channel, and agents never deny being AI when asked.
Can an agent send email as our CEO?
No. "Send as CEO" by agents is forbidden; other send-as uses need explicit delegation and AI-assisted marking.
Does it cover Indian telecom and privacy rules?
An India jurisdiction pack is planned at GA. All rules in it must be verified with counsel.
See Persona Policy Engine in your workflow
Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.