Runtime Agent Governance

The Problem

Your AI agents are making decisions faster than humans can review them. They execute trades, access patient data, modify production systems—and you find out after the fact, from logs written seconds or minutes later.

This is governance theater, not governance.

• Logs tell you what went wrong — but can’t stop it from happening
• Agents share credentials — so you can’t enforce per-agent policies
• Policy lives in code — scattered across services, inconsistently applied
• Regulators expect controls — runtime verification, not forensic reconstruction

When an agent attempts an unauthorized action, you have two choices: stop it at the boundary, or read about it in the logs after the damage is done.

Why This Matters Now

AI agents operate at machine speed. By the time a human sees the log entry, the trade has settled, the PHI was accessed, the system was modified.

• 77% of CISOs say AI adoption is outrunning governance (IBM, 2026)
• MAS SAFR (Singapore Monetary Authority, Jul 2026): Mandates runtime controls that verify and record agent actions before execution—not logs that reconstruct what happened
• J. Willis, “The Commit Boundary”: The industry is shifting from logs (forensic) to controls (preventive)—governance must happen in the path, not after the fact

---

How Affinidi Enables Runtime Governance

Traditional governance happens after action: agents act, logs are written, humans investigate. Affinidi moves governance before action: the Agent Gateway enforces policy at every request, blocks unauthorized actions, and generates cryptographic proof of control.

The shift:

OLD: Agent → Action → Log → (maybe) Alert → Investigation
NEW: Agent → Gateway (policy check) → Authorized Action → Proof

Every agent request passes through the Agent Gateway, which:

1. Verifies agent identity — Who is this agent? Is it authorized to act?
2. Checks runtime policy — What is this agent allowed to do right now?
3. Queries Trust Registry — Has authorization been revoked? Are constraints active?
4. Enforces before execution — Block if policy fails, allow if policy passes
5. Generates cryptographic proof — Immutable record that control was active at decision time

Technical Foundation

• Agent Gateway acts as the control plane: every agent request flows through it, policy is enforced before the request reaches downstream systems
• Policy-as-code: Declarative policies (e.g., “Agent X can read data Y only during business hours and only if two-factor approval was granted in the last 10 minutes”)
• Trust Registry (TRQP): Real-time queries return current authorization status—revocations take effect immediately, not on the next log rotation
• Verifiable identity per agent: Each agent gets a DID, so policies bind to this agent, not a shared service account
• Observable streams: Agent Stream (coming soon) publishes tamper-evident event logs for compliance audit—logs that prove the control was enforced, not just that the action occurred

Technical Deep Dive What are Trust Registries? TRQP lets you query real-time authorization status—'Is this agent still allowed to do this, right now?'—without exposing internal policy logic.

---

Real-World Applications

The same control-before-action pattern works across industries—just different consequences when policy fails.

Financial Services: Trading agents execute orders worth millions. Before Affinidi, there was no pre-flight policy check. With Agent Gateway, every trade is approved by runtime policy before execution, with a full audit trail for regulators.

Healthcare: AI agents access PHI without real-time verification of authorization. Affinidi’s policy engine checks HIPAA-compliant consent before data is pulled. Zero unauthorized PHI access, compliance by design not by audit.

Crypto & Payments: Agents escalate privileges or chain unsafe actions without oversight. Multi-factor approval policies enforced at runtime block escalation at the boundary, not caught in logs.

Enterprise IT: Agents modify production systems with no approval workflow. Policy-as-code defines what each agent can do, enforced before action. No unauthorized changes reach production—downtime prevented, not investigated.

---

Why Affinidi (Not Just Logs)

Control timing matters:

• Affinidi: Before action (runtime enforcement)
• Alternatives: After action (forensic logs)

Policy location matters:

• Affinidi: At the gateway (control plane)
• Alternatives: Inside each agent (scattered, unenforced)

Regulatory alignment:

• Affinidi: MAS SAFR, eIDAS 2.0 compliant
• Alternatives: Audit-focused, not control-focused

Observability:

• Affinidi: Real-time streams + cryptographic proof
• Alternatives: Logs only, no proof of control

Agent identity:

• Affinidi: Verifiable identity per agent
• Alternatives: Generic service accounts

What this means:

• Log-based governance tells you what went wrong after the fact. You can investigate, but you can’t prevent.
• Agent runtimes (AWS AgentCore, Cloudflare Workers for AI) execute agents but don’t enforce centralized, auditable policy at the boundary—policy is scattered inside each agent, inconsistently applied.
• Legacy IAM was built for humans logging into systems with passwords, not for agents making decisions at machine speed with cryptographic identity.

Affinidi provides runtime governance: policy enforcement before action, not forensic reconstruction after. The control happens in the path, not in the log parser.

Proof It Works

• MAS SAFR alignment: Co-authored with HSBC, JPMorgan, Mastercard, Visa, OCBC—runtime controls are the regulatory requirement, not a nice-to-have
• Cryptographic proof of control: The Gateway generates tamper-evident records that the policy was enforced at decision time, satisfying auditors who want proof of preventive controls
• Self-hostable: Deploy the control plane on your infrastructure—regulators care that you control the policy, not a third party

---

Get Started

Whether you’re a developer prototyping policy-enforced workflows, a compliance team evaluating runtime controls, or an enterprise planning governance at scale—there’s a path in.

• Developers — build your first policy-enforced agent workflow in minutes with Agent Gateway, then start coding in the Affinidi Portal.
• Compliance & security — see how runtime policy enforcement works with Trust Registries and TRQP, or book a demo.
• Enterprises — planning governance for hundreds or thousands of agents? Talk to our team about custom policy deployment.

New to the underlying tech? Start with Trust Registries (TRQP), then explore Decentralized Identifiers.

---

Related Resources

Technical Deep Dives

• Trust Registries & TRQP - Real-time authorization queries
• Decentralized Identifiers (DIDs) - How agents get verifiable identity

Related Solutions

• Cross-Org Agent Trust - Agents that transact across company boundaries
• Per-Agent Attribution - Prove which agent did what