AI Use Policy

This policy sets expectations for AI interactions offered by Ordinexis Sdn Bhd through this website and public demos that refer to it. It also describes the review principles we recommend when evaluating iKB, SalamAI, Saloma AI, MataAI and custom AI services.

Customer deployments may add agreed operating rules, approved use cases and controls. Those controls must be evaluated for the actual workflow; a successful public demo does not establish production readiness or regulatory compliance. Read this policy alongside our Terms & Conditions, Privacy Policy and any applicable product agreement.

AI can invent facts or references, misunderstand instructions, omit important details, rely on outdated information, or reproduce bias. Responses may differ when a request is repeated. Fluent language, a realistic voice, a confidence score or a citation is not proof of correctness.

• Check important facts against the original source and confirm that the source actually supports the answer.
• Verify names, dates, numbers, amounts, units, permissions and instructions before taking action.
• Review missing context and conflicting information rather than treating the first response as final.

"AI You Can Audit" and "AI You Can Trust" describe our emphasis on inspectable systems and accountable use, not a promise that every output is accurate. An audit trail records what happened; it does not, by itself, prove that a decision was correct.

• Knowledge AI: an answer can use the wrong document, overlook an exception, or cite an incomplete or outdated source. Check the underlying material and its applicability.
• Live translation: speech recognition and translation can change meaning, especially with names, accents, dialects, background noise, technical language or overlapping speakers. Confirm important statements with the speaker or a qualified interpreter.
• Voice AI: an agent may misunderstand a request, state an incorrect fact or fail to complete an action. Confirm bookings, transactions and commitments in the authorised business system. An AI voice is not necessarily a human representative.
• Vision AI: a camera system can miss an event or raise a false alarm. Camera angle, lighting, obstruction, image quality and sampling affect results. Review the evidence; it is not a substitute for a safety system or human supervision.

Do not use an unreviewed AI output as the sole basis for a decision affecting a person's health, safety, legal rights, employment, access to finance, insurance or essential services. Medical, legal, financial and other regulated matters require appropriate qualified review.

Public demos and website chat are not professional advice, emergency services, certified interpretation, a safety interlock or a substitute for required due diligence. Do not use them for emergency response or situations where a missed or incorrect output could directly cause serious harm.

For a business deployment, assign an accountable owner, define approval and escalation paths, test the failure cases, and provide a way for people to correct or challenge a result. Keep a practical fallback when the AI or its provider is unavailable.

Use fictional or non-sensitive examples. Before using a microphone, read the demo's data-use notice, obtain permission from anyone whose voice may be captured, and understand any recording or transcription settings. Browser microphone permission is not permission from other people in the room.

SalamAI's public browser test and Saloma AI's live Web Talk experience are limited evaluation tools. Do not enter real patient details, banking information, passwords, confidential customer records or private company documents. A demonstration does not authorise real-world transactions or create a service commitment.

Stop the session when you finish and revoke microphone access in your browser if you no longer want it available. Treat any generated conversation, transcription or translation as something to review, not an established fact.

Only provide content that you are authorised to use and share. Obtain the necessary permissions for personal information, recordings, documents, camera feeds, voices and copyrighted material. Do not record or monitor people unlawfully.

Do not assume that a public test has the same data boundary, retention period, training restrictions or confidentiality arrangements as a contracted private deployment. Check the applicable notice and agreement before sending sensitive material.

Make the role of AI clear where a person could reasonably mistake it for a human, an original source or an independently verified conclusion. Do not present generated speech, images, summaries or messages as authentic evidence of an event that did not occur.

• Fraud, deceptive impersonation, scams, fabricated evidence, or misleading representations of another person's identity or consent.
• Unlawful surveillance, unauthorised collection of personal data, stalking, harassment or discriminatory targeting.
• Malware, unauthorised access, attempts to extract another customer's data, or bypassing authentication, safety controls, rate limits or usage restrictions.
• Unlawful or non-consensual calls, recordings, bulk messages or marketing activity. Follow applicable consent, suppression and communications rules.
• Activity that facilitates serious harm, unlawful exploitation, or infringement of another person's rights.

Testing a security issue does not authorise accessing someone else's information. Security audits and probing are not permitted. Follow our Security Audit Policy and report unexpected observations through our Responsible Disclosure Policy.

Define the intended outcome, suitable data, evaluation criteria, human responsibilities and acceptable error rates before production use. Test with representative languages, accents, documents and operating conditions, including cases where the system should decline or hand over.

Use proportionate access controls, monitoring, retention settings and incident processes. Review quality after changes to models, prompts, integrations or source material. Keep operational decisions and corrective action reviewable without collecting more personal data than the workflow needs.

Features such as citations, encryption, redaction, audit records and human handoff depend on the product and configuration. Their presence does not automatically make a deployment compliant with a law or industry standard.

Stop relying on the affected output and arrange human review. If it has triggered a real action, use your organisation's incident or correction process promptly.

Report concerns to hey@ordinexis.com with the product, approximate time, expected behaviour and a redacted example. Do not send credentials, full personal records or confidential recordings in the initial report. We can arrange an appropriate way to investigate.

Ordinexis may restrict a public demo where use presents a safety, security or legal concern. This policy may be updated as products and risks change; the revision date identifies the current version. Contracted deployments remain subject to their applicable agreements.