{"standard":"north-standard-for-ai-gates","version":"1.0.0","publishedAt":"2026-11-11","controlCount":38,"url":"https://www.usenorth.xyz/standard","clusters":[{"name":"Identity & Scope","description":"What the system is, who uses it, and what it touches. These answers determine which additional controls apply.","controls":[{"id":"STD-01","ask":"What stage is this AI use case at?","why":"The stage determines which controls are mandatory — controls are cumulative, each stage inherits everything before it.","goodEvidence":"A one-line statement in the business case naming the stage (e.g. 'promoting from MVP to Production').","sourceControl":"Identify (Use Case) Q1"},{"id":"STD-02","ask":"Who will actually use this system?","why":"Internal staff, customers, or both — this changes the risk surface and the regulatory exposure.","goodEvidence":"The business case names the users and whether they are staff, customers, or partners.","sourceControl":"Identify (Use Case) Q2"},{"id":"STD-03","ask":"Where must the data be stored and processed?","why":"Data residency drives which laws apply and which cloud regions are permitted.","goodEvidence":"The architecture names the region or the residency requirement explicitly.","sourceControl":"Identify (Use Case) Q3"},{"id":"STD-04","ask":"What types of data does the system process?","why":"Data classification drives encryption, access, and logging duties — you cannot protect what you have not classified.","goodEvidence":"A data classification table naming each dataset's class (Secret/Confidential/Internal).","sourceControl":"Identify (Use Case) Q4 + Q7"},{"id":"STD-05","ask":"Which regulations apply to this use case?","why":"Regulatory scope determines which compliance controls are mandatory at the gate.","goodEvidence":"The documents identify which regulations apply (POPIA, GDPR, EU AI Act, FSCA, etc.) and why.","sourceControl":"Identify (Use Case) Q8"},{"id":"STD-06","ask":"What is the design tier of this use case?","why":"The design tier determines the component checklist and which threat mitigations are required.","goodEvidence":"The architecture names the design tier and confirms the component checklist.","sourceControl":"Identify (Use Case) Q9"},{"id":"STD-07","ask":"Which AI models does the system call, and are they pre-trained or custom-trained?","why":"Model scope decides the security requirements that apply (model cards, provider due diligence, fine-tuning controls).","goodEvidence":"The architecture names every model endpoint and whether any model is custom-trained.","sourceControl":"Identify (Use Case) Q10"},{"id":"STD-08","ask":"What is the highest AI agent scope in the use case?","why":"Agent scope determines whether agentic controls and MCP security requirements apply.","goodEvidence":"The architecture names the agent scope (none, simple, or full agentic with MCP servers).","sourceControl":"Identify (Use Case) Q11"},{"id":"STD-09","ask":"How are internal and external integrations identified?","why":"Every integration is a trust boundary — unidentified dependencies are unassessed risks.","goodEvidence":"The architecture or SRD lists all integrations and dependencies.","sourceControl":"Identify (Use Case) Q17"},{"id":"STD-10","ask":"Does POPIA apply to this use case?","why":"South African personal information triggers lawful-basis, minimality and breach duties.","goodEvidence":"The documents identify whose personal information is processed and under what basis.","sourceControl":"Identify (Regulations) Q1"}]},{"name":"Threat & Evidence","description":"The threats the system faces and the evidence that controls are in place. These are what block sign-off when they're open.","controls":[{"id":"STD-11","ask":"How is the use case threat modelled?","why":"A gate without a threat model is a gate without a basis for its controls.","goodEvidence":"A threat model naming threats, rating them, and mapping mitigations.","sourceControl":"Design Q2"},{"id":"STD-12","ask":"How are the Tier 1 OWASP LLM Top 10 threats mitigated?","why":"These are the baseline attacks every LLM system faces — injection, data poisoning, sandbox escape.","goodEvidence":"The threat model maps each OWASP LLM threat to a specific mitigation.","sourceControl":"Design Q3"},{"id":"STD-13","ask":"Are service roles least-privilege, with no wildcard permissions?","why":"Over-privileged roles turn any compromise into a full-tenant compromise — the classic escalation path.","goodEvidence":"IAM policies scoped to specific resource ARNs, no Resource:'*', no role-creation rights on runtime roles.","sourceControl":"Develop Q5"},{"id":"STD-14","ask":"Are model guardrails configured (prompt-attack filtering)?","why":"Documents flowing into models can carry injection; guardrails are the control.","goodEvidence":"Architecture names the guardrail service and the filters enabled (e.g. PROMPT_ATTACK).","sourceControl":"Develop Q6 + Q8"},{"id":"STD-15","ask":"What is logged, and can logs leak personal information?","why":"Logs are the most common accidental PII disclosure path.","goodEvidence":"A logging standard showing what is logged, that payloads are scrubbed, and retention is set.","sourceControl":"Develop Q1 + Q15"},{"id":"STD-16","ask":"Are API endpoints authenticated, authorised, and rate-limited?","why":"Every endpoint is a door — unprotected endpoints allow unauthorised invocation.","goodEvidence":"API specifications naming auth (OAuth/mTLS), authorisation, and rate limiting per endpoint.","sourceControl":"Develop Q11"},{"id":"STD-17","ask":"Are cost controls and denial-of-wallet protections in place?","why":"Repeated or automated prompts can consume tokens and exhaust quotas — DoS by billing.","goodEvidence":"Documentation of cost monitoring, budget alerts, and per-invocation limits.","sourceControl":"Develop Q12"},{"id":"STD-18","ask":"How are secrets managed (API keys, credentials, tokens)?","why":"Secrets embedded in code, environment variables, or logs are exposed secrets.","goodEvidence":"A secrets management process using a vault or managed service, not hardcoded values.","sourceControl":"Develop Q10"},{"id":"STD-19","ask":"How is security scanning built into the CI/CD pipeline?","why":"Vulnerabilities caught in the pipeline are cheaper than vulnerabilities caught in production.","goodEvidence":"CI/CD configuration showing SAST, dependency scanning, and IaC scanning gates.","sourceControl":"Develop Q14"},{"id":"STD-20","ask":"How are the security requirements documented and approved?","why":"A blank security requirements document means no baseline — you cannot gate against nothing.","goodEvidence":"A completed Security Requirement Document, reviewed and signed off before Mobilise.","sourceControl":"Design Q18"}]},{"name":"Operational Floor","description":"Can this system actually run in production? These are the minimum operational capabilities required at the gate.","controls":[{"id":"STD-21","ask":"What are the acceptance criteria for this gate?","why":"The gate must know what 'good enough' was measured against — not a feeling.","goodEvidence":"Named acceptance criteria with results, tied to the stage gate requirements.","sourceControl":"OpRead Q1 + Q2"},{"id":"STD-22","ask":"Is there an incident response plan that covers this system?","why":"When (not if) something goes wrong, the minutes-to-response depend on a plan that exists.","goodEvidence":"A current IRP naming roles, severity ladders, and contacts for this use case.","sourceControl":"OpRead Q3 + Q16"},{"id":"STD-23","ask":"How does a human review low-confidence outputs before they act?","why":"Human-in-the-loop is only real when the threshold, the reviewer, and the override audit are designed.","goodEvidence":"A named confidence threshold, a named reviewer role, and an override log.","sourceControl":"OpRead Q19"},{"id":"STD-24","ask":"How is model drift detected and evaluated?","why":"An AI system that worked at launch can silently degrade — drift detection catches it.","goodEvidence":"Monitoring configuration showing drift detection metrics and evaluation cadence.","sourceControl":"OpRead Q11"},{"id":"STD-25","ask":"How are security threats detected in runtime?","why":"Pre-production controls are not enough — the system must detect threats while running.","goodEvidence":"Threat detection services enabled (e.g. GuardDuty) with alert routing.","sourceControl":"OpRead Q12 + Q13"},{"id":"STD-26","ask":"How are code and infrastructure findings addressed before production?","why":"SAST and IaC findings that are open at the gate are vulnerabilities that ship.","goodEvidence":"Evidence that code scan findings are triaged and remediated or risk-accepted.","sourceControl":"OpRead Q24"},{"id":"STD-27","ask":"How is the workload deployed securely through CI/CD?","why":"Manual deployments bypass controls — the pipeline IS the control.","goodEvidence":"CI/CD pipeline configuration with security gates and approval steps.","sourceControl":"Deploy Q2"},{"id":"STD-28","ask":"How are prompt templates versioned and controlled?","why":"Prompt changes are code changes — unversioned prompts are uncontrolled behaviour.","goodEvidence":"Prompt templates stored in version control with review process.","sourceControl":"Develop Q9"},{"id":"STD-29","ask":"How are data protection measures applied to storage?","why":"Knowledge base and training data buckets need malware scanning and PII controls.","goodEvidence":"S3 bucket policies with encryption, access controls, and malware scanning configured.","sourceControl":"Develop Q10"},{"id":"STD-30","ask":"How is the supply chain assessed for security?","why":"Every dependency is a trust decision — unidentified dependencies are unassessed risks.","goodEvidence":"Dependency inventory with security assessment of critical third parties.","sourceControl":"Develop Q13 + OpRead Q20"}]},{"name":"Testing & Sign-off","description":"How the system is validated before production, and who signs to say it's ready.","controls":[{"id":"STD-31","ask":"What is the testing and evaluation strategy?","why":"Testing proves the system does what it claims — without a strategy, tests are anecdotes.","goodEvidence":"A testing strategy covering functional, security, and evaluation testing.","sourceControl":"Test Q1 + Q2"},{"id":"STD-32","ask":"How are threat model findings validated by test cases?","why":"A threat model without tests is a hypothesis — tests make it evidence.","goodEvidence":"Test cases mapped to threat model findings with pass/fail results.","sourceControl":"Test Q3"},{"id":"STD-33","ask":"Who signs off security test results before production?","why":"Someone must be accountable for the test evidence — a named human, not a process.","goodEvidence":"Named sign-off authority with date and evidence reference.","sourceControl":"Test Q7"},{"id":"STD-34","ask":"How is the threat model revalidated for production?","why":"Threats evolve between MVP and production — the model must be current at the gate.","goodEvidence":"Threat model review date showing revalidation before the production gate.","sourceControl":"OpRead Q23"},{"id":"STD-35","ask":"How is resilience and recovery validated?","why":"A system that cannot recover from failure is not production-ready.","goodEvidence":"Recovery testing results, backup verification, and RTO/RPO documentation.","sourceControl":"OpRead Q25"}]},{"name":"The Gate's Three Asks","description":"The three questions every inherited form forgot. These apply universally, regardless of use case size or risk tier.","controls":[{"id":"GATE-1","ask":"Are quality acceptance thresholds met?","why":"The gate must know what 'good enough' was measured against — the acceptance criteria and their results, not a feeling.","goodEvidence":"Named thresholds with measured results for each acceptance criterion.","sourceControl":"Gate block (H4 finding)"},{"id":"GATE-2","ask":"Is the human-review design confirmed?","why":"Who reviews what, at what confidence threshold, with what override audit — named before go-live, not improvised after.","goodEvidence":"Named reviewer role, confidence threshold, and override audit log design.","sourceControl":"Gate block (H4 finding)"},{"id":"GATE-3","ask":"Is a run-owner named and the operate model funded?","why":"A system promoted to production without a named owner and funded operations is an orphan with uptime.","goodEvidence":"Named owner with budget allocation for ongoing operations.","sourceControl":"Gate block (H4 finding)"}]}]}