Our AI Use Policy

When Mariam Tech delivers AI-driven features (chatbots, document assistants, automation), we clearly label AI-generated content and make best-effort citations or references where applicable. We aim to let users know when a response, summary, or draft was produced by a model rather than a human.

We design and tune AI systems to reduce harmful outcomes and limit bias. Practices include guardrails, periodic bias tests, and avoiding misleading guarantees about business outcomes.

Client and user data are treated as confidential. We adhere to the following principles:

• Purpose limitation: only collect data necessary to provide the service.
• Consent & control: clients can opt in/out of model fine-tuning or analytics.
• Anonymization: pseudonymize/anonymize data where required.

AI helps with drafting, triage, and automation but is not infallible. We therefore:

• Human-in-the-loop: critical decisions (legal, medical, financial) are reviewed by humans.
• Known limitations: models can hallucinate or reflect training gaps.
• Error reporting: clients can report issues and request investigations.

Security controls include encryption in transit/at rest, role-based access, and retention policies with delete/export on request.

Clients can manage data and disclosures:

• Request export/delete: email privacy@yourdomain.com
• Permission to publish: we don’t publish client details without written permission.
• Report an issue: email hello@yourdomain.com

• 2025-09-01 — Clarified consent language for training data.
• 2024-12-15 — Added section on human oversight and bias testing cadence.