AI Disclosure

1. OperelliOS AI Doctrine

OperelliOS is AI-native, but AI is not the source of truth. AI may accelerate work by drafting, extracting, summarizing, classifying, detecting missing information, and recommending next actions.

AI does not have independent authority over money, permissions, compliance, public trust, legal commitments, or irreversible mutations. Deterministic systems and governed human decisions retain authority.

2. Types of AI Assistance

Subject to the features actually enabled, AI-Assisted Features may help users:

• Draft descriptions, summaries, customer communications, and internal notes.
• Extract candidate information from documents into a reviewable draft.
• Classify records or documents and suggest workflow context.
• Identify potentially missing information or inconsistencies.
• Recommend next actions for an operator to review.
• Explain deterministic results in plain language without changing the underlying calculation.

3. Authority Levels

Every AI surface should declare its authority level and failure mode.

4. Accuracy and Human Review

AI output may be inaccurate, incomplete, biased, inconsistent, or outdated. Similar prompts may produce different results.

Users must review and, where necessary, correct outputs before relying on them. Review is especially important for customer-facing content, scopes of work, pricing descriptions, operational instructions, compliance-sensitive information, and any output that may affect a person or business.

OperelliOS does not represent that AI output is complete, unique, or fit for every purpose.

5. Financial and Operational Boundaries

AI may suggest line items, draft scope language, or explain a result, but it must not establish final prices, taxes, totals, balances, payment state, refunds, payroll, profitability, or accounting truth.

AI must not grant permissions, change roles, bypass approvals, approve a public proposal, send funds, create a binding customer commitment, or execute an irreversible action outside the governed mutation path.

The required mutation path is identity, policy, domain mutation, audit, outbox, and controlled side effects.

6. No Professional Advice

AI output is not legal, tax, accounting, insurance, medical, engineering, licensing, or other regulated professional advice.

Users are responsible for consulting qualified professionals where appropriate and for complying with laws and trade requirements applicable to their business.

7. Sensitive Data

Users should not submit health information, government identifiers, financial-account credentials, information about minors, or other highly sensitive personal information to an AI-Assisted Feature unless OperelliOS expressly documents that the specific feature is approved for that information.

The final product should provide just-in-time warnings and technical controls where a workflow is likely to expose sensitive information.

[ENGINEERING VERIFICATION REQUIRED: Confirm field-level data paths, redaction, logging, retention, and provider restrictions.]

8. AI Providers and Data Flow

OperelliOS may use one or more third-party model or infrastructure providers. A provider must not be listed publicly until its production use, purpose, data categories, location, retention, and contract terms are confirmed.

[ENGINEERING VERIFICATION REQUIRED: Confirm AI providers and whether prompts or outputs are retained, reviewed, or used for provider model improvement.]

9. Model Training Policy

Recommended OperelliOS position: identifiable Tenant Data and end-customer information are not used by default to train shared or generally available third-party or OperelliOS models.

Product-improvement analysis may use controlled feedback, evaluation data, or properly de-identified and aggregated information only under an approved policy and technical controls.

Any future program that uses identifiable or tenant-contributed information for model improvement should require separate, informed opt-in; disclose the purpose and retention; allow withdrawal where feasible; and maintain an audit record.

[PRODUCT DECISION REQUIRED: Formally approve no-training-by-default.] [ENGINEERING VERIFICATION REQUIRED: Confirm contractual and technical enforcement.]

10. Document Intelligence

Document workflows should follow: Document → Classification → Extraction Draft → Validation → Human Review when required → Governed Mutation → Audit and Outbox.

AI-extracted content remains a draft until validated. It must not directly create financial truth, invoices, taxes, payments, permissions, or legal commitments.

11. Customer-Facing AI

This Gate 0 disclosure does not authorize an AI receptionist, AI voice agent, autonomous SMS, autonomous chat, call recording, or transcription with real customer interactions.

Those capabilities require separate product terms, communications consent, authority limits, escalation rules, provider disclosure, audit controls, and attorney review before activation.

12. Monitoring, Feedback, and Safe Failure

OperelliOS should maintain prompt and model versioning, traces appropriate to the risk, cost and latency visibility, user feedback, unsafe-action blocking, and a safe fallback when the AI provider is unavailable or confidence is insufficient.

Safe failure means the workflow remains available in deterministic or manual mode, and no hidden or partial mutation is committed.

[ENGINEERING VERIFICATION REQUIRED: Confirm current control-plane implementation before public claims.]

13. Questions

Questions about AI-assisted features or this disclosure: [LEGAL REVIEW REQUIRED: Confirm legal and privacy contact email]

Related documents:

• Terms of Service
• Privacy Policy
• Acceptable Use Policy
• Data Security Overview

**Contact placeholder:** [LEGAL REVIEW REQUIRED: Confirm legal and privacy contact email]