AI Services Terms

These AI-specific terms supplement Iverel's general Terms & Conditions. Where these terms conflict with the general T&Cs in respect of an AI service or AI-augmented deliverable, these terms govern.

1. Models we use

Iverel uses third-party large language models (LLMs) including Anthropic Claude and OpenAI GPT-family models. We use enterprise endpoints with zero-retention policies so that customer prompt and response data is not used to train provider models. Where a deployment has elevated sensitivity (healthcare, legal, finance, or government), we can deploy in a self-hosted configuration so that no customer data leaves the infrastructure agreed in the engagement letter.

2. Data handling and residency

Customer data is processed within Australia or the European Union by default, chosen at the start of an engagement. Data is encrypted in transit (TLS 1.2+) and at rest (AES-256). Credentials and API keys are stored in encrypted secret stores and rotated on agreed cadence. No customer data is shared with any third party outside the model provider chain agreed in the engagement letter.

3. Accuracy and probabilistic output

AI outputs are probabilistic. Iverel does not warrant that any individual output is correct, complete, or fit for any specific purpose without human review. Every production system Iverel ships includes confidence-thresholded gating: high confidence actions execute automatically, lower confidence actions are routed to a human reviewer in the customer's organisation. The customer is responsible for staffing those reviewer roles unless explicitly contracted otherwise.

4. Human-in-the-loop responsibility

Decisions with material legal, financial, or safety consequences are not made autonomously by Iverel-built systems unless explicitly authorised in writing by the customer's nominated executive sponsor. The customer agrees that final responsibility for any decision the AI participated in, whether reviewed by a human or not, rests with the customer.

5. Drift, retraining, and tuning

Models drift. Provider model versions change. Integrations evolve. Iverel's monthly operations fee covers prompt tuning, threshold retuning, and integration health checks for the systems within scope. Without an active operations engagement, Iverel does not warrant continued accuracy of a delivered AI system beyond the post-launch monitoring window stated in the engagement letter.

6. Intellectual property

On full payment, the customer owns the workflows, prompt assets, and any custom code Iverel builds for them. Iverel retains ownership of internal tooling, base agent frameworks, and re-usable components used across customers. Where Iverel uses an open-source component, the component is licensed under its respective terms. Iverel will provide a software bill of materials on request.

7. Logging, observability, and audit

Every action taken by an Iverel-built AI system is logged with input, output, model version, decision threshold, and any human review action. Logs are retained for the period stated in the engagement letter (default 90 days for production systems). The customer can export logs at any time, in standard JSON format.

8. Hallucination and source-grounding

For systems that produce factual claims (research assistants, customer support agents, document analysers) Iverel implements retrieval-augmented generation (RAG) so that responses are grounded in the customer's authoritative sources rather than the model's training data. Iverel does not warrant that grounding eliminates hallucination, but designs and tests systems to keep hallucination rate below the threshold agreed in the engagement letter.

9. Liability

Iverel's aggregate liability for any claim arising from an AI service is limited to the fees paid by the customer for that service in the twelve months preceding the claim. This limit applies in addition to any limits in the general T&Cs and is the maximum exposure across both documents.

10. Changes to these terms

Iverel may update these AI Services Terms from time to time to reflect changes in the underlying model providers, regulatory requirements, or our internal operational standards. Material changes will be notified to active customers in writing at least 30 days before effective date.

Questions about these terms? Email [email protected]. For commercial enquiries, see /pricing.