Responsible AI & Disclosures负责任 AI 与信息披露

1. What our AI services are

Telixus provides three categories of AI-related services:

• AI application development — we design and build AI software for clients under service agreements.
• AI routing gateway — the Telixus.Co platform provides unified API access to AI models operated by third-party upstream providers, with routing, billing, quota and security controls.
• AI security services — evaluation (red teaming, benchmarking, compliance gap analysis) and protection (guardrails, filtering, monitoring) for LLM applications and AI agents.

Telixus does not train or operate its own foundation models. When you use the gateway, model responses are generated by the upstream provider you selected.

2. Third-party models and trademarks

The gateway routes requests to models operated by providers such as Anthropic, OpenAI and Google. These providers apply their own terms, usage policies, safety systems, regional availability and rate limits to requests, and those constraints flow through to gateway users.

Anthropic, Claude, OpenAI, GPT, Google, Gemini and other product names referenced on this site are trademarks of their respective owners. Telixus is an independent company. We are not affiliated with, sponsored by, or endorsed by any upstream model provider, and nothing on this site should be read as implying such a relationship.

3. Limitations of AI outputs

AI-generated outputs — whether produced through our gateway, inside applications we build, or in our own tooling — have inherent limitations. They may be:

• Inaccurate or fabricated ("hallucinations"): fluent, confident text that is factually wrong;
• Incomplete or outdated: models have knowledge cutoffs and may miss recent events;
• Biased: outputs can reflect biases present in training data;
• Inconsistent: the same input can produce different outputs across runs or model versions;
• Manipulable: adversarial inputs (e.g. prompt injection) can cause unintended behavior.

No output from an AI system should be relied upon without appropriate verification proportionate to the stakes of the decision.

4. Human oversight requirements

Consistent with our Terms of Service, AI outputs must not be used as the sole basis for legal, medical, financial, safety-critical, employment, credit, housing, insurance or other high-impact decisions without qualified human review and any legally required safeguards.

In systems we build for clients, we recommend — and where engaged for security, we implement — human-in-the-loop checkpoints for irreversible or high-impact actions taken by AI agents.

5. Data handling and model training

• We do not train foundation models, and we do not use your prompts or outputs to train any Telixus model.
• Raw content logging is off by default on the gateway: prompts and responses are not stored in ordinary logs. Content logging can only occur in narrowly scoped situations (e.g. abuse investigations, user-requested diagnostics, or configurations an organization's administrator explicitly enables), with access controls and limited retention.
• Content must reach the upstream provider to be processed. Your prompts, inputs and files are transmitted to the third-party model provider you selected, and are handled under that provider's terms and privacy policy.
• Operational metadata is retained (request IDs, model, token counts, latency, cost, error categories) as needed for billing, security and troubleshooting.

Full details, including your rights, are in the Privacy Policy.

6. Security practices

We apply administrative, technical and organizational measures across our platform, including access controls, hashed storage of downstream API keys, credential redaction, encryption in transit, audit logging, rate limiting and abuse monitoring. Our AI security work — for our own platform and for clients — is aligned with recognized frameworks including the OWASP LLM Top 10, NIST AI Risk Management Framework, MITRE ATLAS and EU AI Act readiness practices.

No system is absolutely secure. We encourage responsible disclosure of vulnerabilities to support@telixus.co.

7. Prohibited uses

Our services must not be used to create, transmit or support unlawful, harmful, abusive, fraudulent or infringing activity — including malware generation, phishing, unauthorized access, exploitation of minors, targeted harassment, or evasion of sanctions and export controls. The complete list is in Terms of Service §4. We enforce these rules through automated and human review, and we suspend keys and accounts that violate them.

8. About our security assessments

Telixus security evaluations are professional assessments producing evidence-based reports. Unless explicitly stated in an engagement, they are not certifications and do not constitute accreditation by any government or certification body. Assessment results reflect the system tested, at the time tested, under the scope agreed — AI systems change, and periodic re-evaluation is recommended.

We conduct offensive security testing (red teaming) only with the authorization of the system owner and within an agreed scope.

9. Website content disclosure

Some text, illustrations and code samples on this website may have been produced with AI assistance and reviewed by our team. Content on this site is for general information only and does not constitute legal, financial, medical or other professional advice, nor an offer capable of acceptance. Product capabilities described here may evolve; the authoritative description of the platform's obligations is in the Terms of Service.

10. Reporting and contact

• Abuse or misuse of our platform: support@telixus.co
• Security vulnerabilities (responsible disclosure): support@telixus.co
• Privacy requests: support@telixus.co — see the Privacy Policy for your rights
• Business inquiries: admin.sg@telixus.co

We may update this statement from time to time; the effective date above will change when we do.