Read this as an informative engineering correspondence, not legal advice or a certified conformity assessment. The CIRIS Constitution itself marks these article mappings “pending legal review.” Every mechanism below was verified live in code by a skeptic whose job was to refute it, but the question of whether that satisfies the law is one for your counsel, not this page.
| Article | What it requires | How CIRIS addresses it |
|---|---|---|
| Article 9 | Run a continuous, whole-lifecycle risk process: identify, evaluate, and mitigate foreseeable risks, with particular weight on vulnerable groups and, per 9(7), on training-data bias. | Every action passes a five-stage decision pipeline (PDMA / CSDMA / DSDMA / IDMA / action-selection) plus a conscience layer before it executes, with a hard-coded DISCRIMINATION prohibition and a signed audit trail as the record that measures worked. For extreme-tail risk, the Constitution's Catastrophic-Risk Evaluation gate requires an independent red-team and dual external sign-off. |
| Article 12 | Automatically log events over the system's lifetime so behavior is traceable. | A signed, hash-chained audit trail (SHA-256 chained, Ed25519-signed) rather than a plain log file, so every recorded action is tamper-evident and independently verifiable. |
| Article 13 | Operate transparently enough that a deployer can interpret the output and use the system correctly. | The PDMA carries a continuous-monitoring publication duty, and the reasoning-trace API exposes the decision's stage IDs and a contestability path, so the reasoning behind an action is inspectable rather than opaque. |
| Article 14 | Design for effective human oversight, with a human able to intervene in or stop the system. | The A0 to A4 autonomy scale grades every deployment by who bears the action, Wisdom-Based Deferral routes uncertainty to a named human Wise Authority, and the fail-secure kill switch lets named humans halt the agent. |
| Article 15 | Be accurate, robust, and cybersecure across the lifecycle, resilient to error and adversarial manipulation. | Hybrid post-quantum signing (Ed25519 + ML-DSA-65) with 2-of-3 fail-secure multi-source verification, the Constitution's adversarial-robustness playbook, and measured resilience: zero cross-group leakage across 400 nodes, and the corpus reconstructing with a fifth of holders offline. |
| Article 16 | Providers of high-risk AI must ensure oversight, logging, and conformity are in place throughout. | A pre-execution conscience gate on every action, plus the higher autonomy tiers that push each decision's stage ID, verdict, and risk band to a human oversight surface. |
| Article 50 | Tell people when they are interacting with AI, and mark AI-generated content machine-readably (applies from August 2026). | AI self-disclosure ships today: is_bot flags and hard-coded, locale-enforced disclosure on every message adapter. Machine-readable content marking via C2PA Content Credentials is being adopted at the media boundary for the August 2026 deadline, so this article is one limb shipped, one limb adopting. |
| Article 72 | Actively monitor the system in the field after deployment and feed issues back. | The Constitution's continuous-compliance loop runs telemetry streams into drift detection into review, a standing monitoring loop rather than a one-time audit, surfaced through CIRISLens observability. |
Mapped against the canonical CIRIS Constitution as of July 2026. Informative correspondence, pending legal review. Tell us if a mapping is wrong and we will fix it.
Sources
- Article 9CIRISAgent DMA pipeline + Constitution Annex D (CRE)
- Article 12CIRISAgent SignedAuditService (hash chain + signing)
- Article 13CIRISAgent reasoning-trace API + Constitution PDMA Step 6
- Article 14CIRISAgent DeferHandler + Constitution CC 7.5.3.1 autonomy tiers
- Article 15CIRISVerify PQC signer + CIRISServer measured robustness
- Article 16CIRISAgent conscience gate + defer routing
- Article 50CIRISAgent adapter disclosure + Constitution C2PA profile
- Article 72CIRISLens monitoring + Constitution Annex H
The same mechanisms mapped to the security taxonomy, and the whole accountability landscape.