AI Governance & Safety

SIFTR is built on the principle of Deterministic Auditing. We prioritise accuracy over generative creativity.

🚫 No Hallucinations

The SIFTR engine is programmatically restricted to Quote-Only output. It is forbidden from paraphrasing, “summarising,” or creating content not present in the source PDF.

🛡️ Model Privacy

• No Training: We do not use customer data to train, fine-tune, or improve any AI models.
• Statelessness: Documents are processed in memory and are not retained for model improvement.
• Sovereign Only: All AI processing occurs within AWS London (eu-west-2) infrastructure.

⚖️ Bias Mitigation

SIFTR audits against the NCSC CAF 4.0 Standard, not subjective criteria. By using the official IGPs as the logic anchor, we ensure every assessment is objective and repeatable.