Magenta Canon is a runnable evaluation context for legal decision systems: it models authority binding for legal AI recommendations, disclosure requirements, verification gates for high-stakes legal actions, and an append-only, tamper-evident audit chain. It does not govern legal substance, case strategy, or attorney-client privilege, and it is not a certified legal-compliance offering.
- Authority binding: legal AI recommendations require explicit grant before execution
- Verification gates: high-stakes actions blocked pending human approval
- Audit chain: append-only, tamper-evident record of every modeled legal AI decision and its authority source
JURISDICTION
Law & AM100 Firms
The accountability gateway applied to legal decision systems. Authority binding, verification gates, and audit chains for regulated legal workflows.
How to Read This Proof Room
Four states, kept apart.
- Runnable / proven control pattern — executes today against the gate: delegated actions are allowed, withheld actions are blocked before downstream, both are recorded, and the evidence verifies independently.
- Reference / mock integration — the downstream tool in these examples is an audited mock, not a live provider. It proves the control pattern and the downstream absence of a blocked call, and nothing about a real provider's behavior.
- Future provider integration — connecting a real provider is design-partner work that has not been built. No live provider connection exists, and none is implied by a workflow appearing on this page.
- Not claimed — no compliance certification, no professional advice, no real-world side effects, and no assertion that a customer's own policies are correct.
Enforcement is scoped to the MCP tools/call method class; other MCP methods pass through the gateway ungated. The refund ceiling is the one numeric threshold the gate enforces today — other boundaries are expressed as tool-scoped allowlists under default-deny.
What Magenta Governs
Against audited mock tools, the gate enforces delegated versus withheld authority for legal AI workflows. Runnable today: delegated legal research and memo drafting proceed, while court e-filing is blocked at the gate and never reaches the downstream tool.
- Authority binding — a legal AI action executes only under an explicit operator-delegated grant (runnable)
- Withheld high-stakes actions — filing and other irreversible steps are refused before downstream (runnable, against mock tools)
- Evidence — every allow and block becomes a signed, hash-chained receipt an independent verifier can check (runnable)
- Human intervention — witnessed and fail-closed; it halts or refuses rather than opening the gate. A fully wired production override channel is roadmap
- Live practice-management or court-system integration — future design-partner work, not built
What Magenta Does Not Govern
Magenta Canon is an accountability gateway for AI-agent tool calls, not a legal services provider. The following fall outside the scope of the gateway.
- Legal substance or correctness
- Case outcomes or strategy
- Attorney-client privilege
- Statutory interpretation
Related Governance Jurisdictions
The Magenta Canon accountability gateway also applies to financial systems, healthcare, and autonomous agents — see the full jurisdictions overview.
Clarification: This refers to enforcing governance rules in legal contexts, not providing licensed legal advice. Domain references describe governance rule contexts, not licensed legal, financial, or medical services.