Guardrails & Filtering

Design the system prompt architecture with privilege separation and trust tier definitions at design stage.

Implement input sanitisation and injection detection filters covering known injection patterns and privilege escalation attempts.

Wrap all untrusted content in random delimiters and datamarking; instruct the model never to execute instructions inside the marked region. Gate release on injection eval results.

Benchmark the classifier on a labelled injection corpus and tune the decision threshold. Sign off the operating point before deployment.

Define content safety policy at use case design stage. Classify prohibited content types and set zero-tolerance thresholds.

Select a foundation model with documented RLHF or Constitutional AI safety training. Verify against toxicity benchmarks.

Implement multi-layer content moderation (input + output) validated against toxicity benchmarks. Escalate when filter bypass rates spike.

Maintain live HITL review for deployments serving vulnerable users or high-risk contexts. Escalate confirmed toxic outputs immediately.

Design system prompts to explicitly prohibit toxic, hateful, and harmful content generation.

Establish data transfer and storage policy for AI training data. Enforce approved storage locations from point of collection.

Implement DLP controls in the data acquisition environment to prevent unauthorised extraction or transfer of training data.

Enforce data handling policy in the build environment. Require explicit approval for any data transfers outside the environment.

Configure DLP controls in the build environment to block training data from leaving approved boundaries.

Conduct a privacy risk assessment at use case design stage. Determine if a DPIA is required before data acquisition.

Apply S1-defined privacy controls during data acquisition: verify consent, minimise data, anonymise personal data.

Apply anonymisation and masking controls to personal data before use in model training. Validate de-identification effectiveness.

Apply Privacy by Design in model architecture using differential privacy or federated learning where technically feasible.

Publish the privacy notice and confirm consent management is operational before go-live.

Define and sign off a purpose-to-data-source matrix with lawful basis at intake. Make it the approved baseline for runtime enforcement.

Sign zero-retention/no-training terms with each model provider and obtain DPO sign-off on the data flow before enabling any endpoint.

Apply robust de-identification (k-anonymity, l-diversity, differential privacy) during data processing. Validate effectiveness.

Implement output filters to detect and suppress quasi-identifying attribute combinations in model responses.

Propagate source ACLs and classification labels onto every chunk at ingestion. Reject documents whose entitlements cannot be resolved.

Scan every model response inline with DLP before delivery; redact or block PII, PAN and MNPI matches. Keep the rule set version-controlled.

Implement confidence scoring to communicate output certainty alongside each result. Calibrate before deployment.

Define model accuracy acceptance criteria aligned to business requirements before validation commences.

Implement counterfactual explanation to show users what changes would alter the model's output.

Communicate model accuracy, known limitations, and uncertainty to users in the production interface at launch.

Monitor production accuracy continuously against the validated baseline. Trigger model review when accuracy degrades.

Specify a RAG architecture at design stage for factual domains. Define grounding requirements and acceptable hallucination thresholds.

Evaluate foundation model candidates on hallucination benchmarks at design stage. Select models with lowest documented rates.

Design system prompts to instruct the model to acknowledge uncertainty, cite sources, and refuse when knowledge is insufficient.

Fine-tune on a curated, domain-specific dataset to improve factual accuracy. Validate hallucination rates pre/post fine-tuning.

Configure conversation controls at deployment to restrict the model to approved topic domains and escalate off-topic queries.

Establish acceptable hallucination rate thresholds and grounding requirements as policy before build. Assign a named risk owner.

Configure tiered HITL review for high-stakes factual outputs with defined trigger criteria and reviewer SLAs.

Calibrate the initial entropy threshold on a knowledge-boundary dataset; approve sampling design and thresholds per risk tier.

Map each fact class to a designated tool, embed the no-ungrounded-assertion prompt, and gate build review on grounding tests passing.

Resolve every emitted citation against the approved corpus and verify span-level entailment before display. Strip or withhold claims with fabricated or non-entailing references.

Define and approve the source allow-list and write-time scanning during build. Prove non-allow-listed and injection-bearing writes are rejected before go-live.

Conduct a prompt injection threat assessment at design stage covering all input vectors (user, tool, external data).

Penetration test all prompt injection pathways in the system. Prioritise external tool and document ingestion channels.

Build the versioned injection corpus into CI/CD as a pre-release gate. Baseline attack success and sign off the release threshold.

Prioritise jailbreak and adversarial safety testing in pre-deployment validation. Block deployment if prohibited outputs pass filter.

Conduct targeted red team exercises to elicit toxic outputs through jailbreaks and adversarial prompts. Treat bypass as blocking defect.

Monitor production for anomalous data transfers in real time. Alert on any transfer outside approved data flow boundaries.

Tag personal data with subject identifiers at ingestion and maintain an artefact inventory map of every store it reaches. Keep lineage current so erasure can propagate.

Seed registered canary records into the fine-tuning corpus during data preparation. Control the seed manifest so canaries stay traceable and tamper-proof.

Define and execute a domain-specific hallucination test suite before deployment. Treat hallucination rate above threshold as a blocking defect.

Construct synthetic evaluation datasets for knowledge-boundary scenarios. Use to validate model refusal behaviour.

Calibrate the groundedness threshold against the hallucination test suite pre-release; sign off the threshold in the validation pack.

Verify a signed attestation and content hash on every dataset shard at ingestion. Reject unsigned or hash-mismatched data before it reaches the training pipeline.

Gate every model promotion on backdoor-trigger probes and a behavioral diff against the approved baseline. Block release on significant regressions or trigger-pattern anomalies.

Conduct comprehensive prompt injection red team exercises (direct, indirect, multi-turn) before deployment.

Classify content sources into trust tiers at design; place privileged tools behind a tier requiring user-originated intent or human approval. Sign off the trust-tier map before build.

Re-run injection evals on every template change and periodically against new attack techniques. Manage the spotlighting wrapper under change control.

Use user feedback, reviewer escalations, and monitoring signals to identify and remediate content safety gaps iteratively.

Monitor for privacy incidents in production including personal data appearing in outputs. Notify regulators within required timeframes.

Tag every memory and vector record with subject-id and retention class; partition stores per tenant/user. Prove the erasure and isolation paths in testing before release.

Penetration test AI system data access boundaries (API endpoints, system prompt exposure, memory leakage).

Conduct periodic data leakage audits including training data memorisation testing. Escalate confirmed leakage incidents to PDPA notification process.

Implement tamper-evident capture of prompts, outputs, and version state during build. Verify a full incident timeline can be reconstructed before go-live.

Run agent tool calls in a network-restricted sandbox behind a deny-by-default egress allow-list. Require security approval for any destination added.

Use production feedback (user corrections, fact-check failures) to drive periodic RLHF cycles. Update model when error rates trend upward.

Require user-facing interfaces to disclose Gen AI limitations and hallucination risk before go-live.

Score every RAG answer for groundedness before release; block, fall back, or escalate responses below the faithfulness threshold.

Sample multiple generations for high-stakes queries and abstain, fall back, or escalate when semantic entropy exceeds the calibrated threshold.

Scan every ingestion batch with spectral-signature and clustering detectors before training. Quarantine flagged clusters for human review against documented thresholds.