START HERE — Native M5 Agents in a Member-Controlled M5POD

[satonakaoshimoto]

Native M5 Agents in a Member-Controlled M5POD

Status: Public architecture and participation discussion. This thread describes reviewable M5 concepts and invites evidence, objections, use cases, threat analysis, and proposed improvements. It does not represent that every described agent or workflow is deployed, credentialed, certified, or authorized for production use.

The M5 model begins with a named, accountable human or lawful entity—not a wallet, model, API key, device, or autonomous agent.

M5HUM / lawful legal principal
        ↓
account context + current role + credential + jurisdiction
        ↓
explicit, revocable delegation
        ↓
M5Canon Six-Gate evaluation
        ↓
M5AGT bounded action
        ↓
authorization / evidence receipt

M5Canon is not an AI agent. It is the deterministic policy and authority-control layer that checks whether a consequential action may activate. M5AGT is the canonical type for an autonomous/software agent acting under bounded delegated authority.

See the architecture

These are Illustrative Public Review Drafts. Maintained repository text and governance records control where visual shorthand differs.

Three dimensions that must remain separate

1. Who the agent serves — account context

• 000-IAM — Identity & Access Management
• 001-M5BOM — Bank of Me
• 002-M5BOU — Bank of Us
• 003-M5BOB — Bank of Business
• 004-M5BOI — Bank of Institutions
• 005-M5BOG — Bank of Government / Governance

2. What right or economic state is involved

• M1 — Money + Utility
• M2 — Commodities
• M3 — Titled / Unique / Collectible RWA
• M4 — Financial Wrapper
• M5 — Jurisdictional Security-State Profile

3. What the agent may do now

Principal identity, current role and credential, jurisdiction, deterministic policy, required accountable approval, and evidence/audit anchoring must all be evaluated. An internal M5 label never creates a legal conclusion, government authority, regulated status, or transfer right.

What we want participants to help organize

1. Native-agent identity, provenance, role, and lifecycle.
2. Account-context boundaries from Bank of Me through shared, business, institutional, and government/governance contexts.
3. M5Canon policies, approvals, revocation, receipts, and conformance tests.
4. M5Brain and sovereign-graph context, memory, relationships, and provenance.
5. Threat vectors: prompt injection, impersonation, confused deputy, stale grants, tool escalation, leakage, recovery abuse, and compromised devices or agents.
6. Authenticated communications, notices, consent, disclosure, and agent-to-agent handoffs.
7. Money, assets, title, financial wrappers, security-state profiles, and external legal/regulatory classifications.
8. Jurisdiction, legal authority, due process, professional credentials, and accountable providers.
9. Local/offline models, external models and APIs, portability, provider substitution, and no-lock-in requirements.
10. SOPHIA roles, M5-CV capability evidence, training, testing, mentoring, and human-reviewed activation.

Public/private boundary

Please contribute schemas, test cases, threat analysis, diagrams, policy questions, public-safe workflows, and interoperability requirements. Do not post private M5POD contents, prompts, agent memory, identity documents, credentials, keys, wallet or account information, personal communications, protected records, confidential business processes, or unreported vulnerabilities.

Public discussion, a role label, successful authentication, or an agent configuration does not create authority. Security vulnerabilities must be reported privately through the repository's security process.

Read first

• M5AGT Credentialed Operations
• M5Canon Six-Gate Authority and Transfer Control
• M5 Economic Architecture
• M5POD Member Activation Architecture
• M5POD RAG Authorization Threat Model

Start below: Which agent capability, account context, threat, workflow, legal boundary, or conformance test can you help define or review?

M5-agent review series

1. Account contexts — Bank of Me through Bank of Government / Governance
2. M5Canon Six-Gate controls
3. M5Brain, sovereign graphs, memory, context, and provenance
4. Threat vectors and fail-closed controls
5. Agent communications, authenticated notice, consent, and handoffs
6. Account, money, asset, financial-wrapper, and security-state classification
7. Legal and jurisdictional authoritative sources
8. Native agents, external models, tools, and provider substitution
9. SOPHIA roles, agent teams, training, and verification
10. Agent lifecycle, receipts, conformance, and recovery

For company, dataset, product, workforce, and training proposals that extend beyond agent architecture, use Discussion 22 — Build the Public Economic Data Commons.