Know Your Agent (KYA) v1.1

Technical Proposal by Lifetime Oy
Lead Architect: Risto Anton
Status: Implementation Ready (Q1 2026 Launch)

1. Executive Summary

In the 2026 industrial landscape, identity verification is no longer a one-dimensional problem. Traditional KYC (Know Your Customer) identifies the human at the door, but fails to govern the autonomous entities they bring with them.

As enterprises integrate "Bring Your Own" (BYO) agents — ranging from Microsoft CoWork and Tencent Workbuddy to proprietary industrial tools like Clawbot — a new governance layer is required.

Lifetime Oy introduces the KYA (Know Your Agent) Standard: a hardware-level framework for anchoring autonomous intent to verified identities through MicroVM isolation and forensic observability.

Era	Identity Problem	Solution
2010s	Who is the human?	KYC — document verification, AML screening
2020s	Who is the business?	KYB — corporate registry, UBO checks
2026+	Who controls the agent?	KYA — identity attribution + capability gating + forensic audit

2. The Three Pillars of KYA

Identity Attribution — The Anchor

Every agent session is cryptographically anchored to a KYC-verified human or entity. This ensures legal accountability for autonomous actions within the DWS10 Control Room.

Agent Action → Session Token → Verified Identity → Legal Entity

Capability Gating — The Leash

Using real-time risk telemetry (via API partners), DWS IQ dynamically adjusts an agent's permissions. If a user's risk profile fluctuates, the agent's compute, memory, and API access are throttled in milliseconds.

Forensic Observability — The Firehorse Audit Trail

The Lifetime Firehorse product line serves as the ecosystem's tamper-proof "flight recorder." It captures a sub-second telemetry stream of every syscall and network request, providing an immutable audit trail bridging agent behavior to human identity.

3. Technical Architecture: Hardware-Level Isolation

To meet the "Zero-Trust" requirements of industrial-grade environments, DWS IQ enforces physical separation for every agent deployment.

Firecracker MicroVMs

Every BYO process is encapsulated in a dedicated MicroVM. No shared kernel access. Each sandbox runs its own isolated guest kernel, eliminating "breakout" exploits and ensuring that a compromise in one agent cannot migrate to the host or adjacent sandboxes.

Property	Value
Boot time	< 125 ms
Memory overhead	< 5 MB per MicroVM
Guest kernel	Isolated per sandbox
Host kernel access	None (paravirtualized)
Storage	Ephemeral by default, persistent opt-in

Hypervisor-Level Egress Control

Network traffic is gated at the VMM (Virtual Machine Monitor) layer. Agents have zero outbound access unless the KYA profile explicitly whitelists an endpoint for the specific session.

Operator (KYA Score: 87)
    |
    +-- MicroVM-1: CoWork      [egress: microsoft.com]
    +-- MicroVM-2: Workbuddy   [egress: api.tencent.com]
    +-- MicroVM-3: Clawbot     [egress: none (air-gapped)]
    +-- MicroVM-4: Cowork      [egress: api.anthropic.com]

Cross-sandbox access? --> TRAP --> SESSION FROZEN

4. Agent Trust Score (ATS)

Formula: ATS = max(0, 1.0 − V×0.25 − L − D×0.05)
Where V = KYA violations, L = latency penalty (0.10 if avg > 500ms), D = denied path attempts.
Hard kill: ATS = 0.0 if manifest is unsigned or TTL expired.

ATS	KYA Score	Permission Level	Decision
≥ 0.80	≥ 80	Full sandbox — egress, filesystem, all MCP tools	GO
0.50–0.79	50–79	Restricted sandbox — no egress, read-only FS	GO (limited)
< 0.50	< 50	Suspended — human review required	NO_GO
= 0.00	= 0	Frozen — agent terminated, investigation	NO_GO

Implementation: src/identity/ats-engine.ts — evaluated before external IDV calls (Shufti Pro) to save API cost on low-trust sessions. 4 violations = zero trust.

5. Use Case: The Multi-Agent Control Room

A Tier-1 client initiates a complex task in the Control Room (managing 16 proprietary industrial models). The client simultaneously deploys four heterogeneous agents:

#	Agent	Role	Provider
1	Microsoft CoWork	Documentation / PM	Microsoft
2	Tencent Workbuddy	Supply-Chain APIs	Tencent
3	Clawbot	Mechanical Optimization	OpenClaw
4	Anthropic Cowork	Strategy / Logic	Anthropic

Interdependency Enforcement: If the Tencent agent attempts to access a resource reserved for the Clawbot, the DWS IQ hypervisor traps the command and freezes the session instantly. KYA score impact: −30 points, operator notified via webhook and dashboard alert.

16 Industrial Models

Each model runs in its own sandbox within the Control Room, serving one of the 16 EU-regulated industries under Fit for 55:

Power & Heat

Iron & Steel

Cement & Lime

Chemicals

Aviation

Maritime

Road Transport

Construction

Agriculture

Waste Mgmt

Aluminium

Pulp & Paper

Petroleum

Food & Bev

Glass & Ceramics

Mining

6. BYO Agent Registration Flow

Stage	Gate	Blocking?
1. Registration	Agent manifest submitted	—
2. Identity Binding	IDV — verify human	Yes
3. Org Verification	KYB — verify organization	Yes (BYO agents)
4. Capability Review	MCP tool approval	Yes
5. Sandbox Entry	KYA score ≥ 50	Yes
6. Runtime	Continuous telemetry + webhooks	Dynamic

7. Compliance Alignment

Regulation	KYA Coverage
EU AI Act (Article 12)	Forensic logging, human oversight, traceability
EU AI Act (Article 14)	Human-in-the-loop via KYA score thresholds
NIS2 Directive	Supply chain security for BYO agents, incident reporting
GDPR	Data residency (EU-only), right to erasure (agent data)
Fit for 55	16 industrial models comply with ETS/CBAM/CSRD
DORA	ICT risk management for financial sector agents

8. Subagent Registry v1.1

Every subagent initialized within the DWS IQ platform must register with the Subagent Registry and sign a local Capability Manifest before execution begins. This ensures that testing and sandbox environments enforce the same identity-gated access controls as production.

Capability Manifest

Upon initialization, each subagent signs a Capability Manifest declaring its permitted resource paths. The manifest is immutable for the duration of the session.

Path	Access	Description
`/mock-api/`	Read/Write	Mock API endpoints for testing
`/temp-db/`	Read/Write	Temporary database for sandbox operations
`/prod-secrets/`	DENIED	Production secrets — access triggers KYA Violation

KYA Violation: Unauthorized Path Access

Enforcement: If a subagent attempts to access a denied path (e.g., /prod-secrets/), the sandbox monitor intercepts within < 1 ms, logs a KYA Violation to the Firehorse audit trail, terminates the session within < 50 ms, and reduces the KYA Trust Score by 40 points. The operator is notified via webhook and dashboard alert.

Step	Action	Latency
1	Path access intercepted by sandbox monitor	< 1 ms
2	KYA Violation logged to Firehorse audit trail	< 5 ms
3	Subagent session terminated	< 50 ms
4	Operator notified (webhook + dashboard alert)	< 1 s
5	KYA Trust Score reduced by 40 points	Immediate

Manifest Schema

{
  "manifest_version": "1.1",
  "subagent_id": "<uuid>",
  "session_id": "<kya-session-token>",
  "signed_at": "<ISO-8601>",
  "permitted_paths": ["/mock-api/", "/temp-db/"],
  "denied_paths": ["/prod-secrets/"],
  "ttl_seconds": 3600,
  "kya_score_at_sign": 85
}

Fault Attribution (Liability Mapping)

When a subagent produces an incorrect result or triggers a violation, the system determines which component is at fault. Fault attribution records are stored separately in Supabase (kya_fault_attribution) — not in the Firehose audit trail — because faults need mutable state (resolved/unresolved) and real-time dashboard queries.

Fault Class	Responsible Component	Example
`LOGIC_FAULT`	Logic Subagent	Incorrect calculation, hallucinated data, business rule violation
`INTEGRATION_FAULT`	Integration Mock/Adapter	Stale mock data, API timeout, schema mismatch
`MANIFEST_FAULT`	Capability Manifest	Manifest too permissive or missing deny rules
`OPERATOR_FAULT`	Human Operator	Misconfigured agent parameters, wrong model selected

Why Supabase? Firehorse is an immutable, append-only forensic log (“what happened?”). Fault attribution is an operational concern (“who fixes it?”) that needs mutable state, RLS, real-time subscriptions, and queryable JSONB evidence payloads.

9. Identity Provider Integration

The KYA architecture is designed for pluggable identity providers. Any IDV/KYC partner that can deliver real-time verification scores and continuous risk webhooks can serve as the Identity Anchor.

KYA Layer	Required Capability	Protocol
L1 — Identity Attribution	Real-time IDV (document + face match + AML)	REST API, < 30s response
L2 — Agent Provenance	KYB (business/organization verification)	REST API
L5 — Continuous Risk	Risk change notifications	Webhook (push), < 5s delivery

Every BYO agent deployed in the DWS10 ecosystem generates a verification event. With 20 industries and multi-agent sessions, this creates volume at scale — not one-off KYC checks.

10. Agentforce Integration — Salesforce Headless 360

Salesforce Headless 360 exposes the entire Salesforce, Agentforce, and Slack platform as APIs, MCP, and CLI. DWS IQ is built the same way. The integration is one MCP handshake.

KYA becomes the EU AI Act compliance layer for every Agentforce agent handoff — covering the gap that base Agentforce does not close natively.

Agentforce Moment	KYA Enforcement	EU AI Act Coverage
Agent spawned	KYC-verified human identity bound to session	Art. 13 — transparency
Tool call issued	Capability Manifest checked at VMM layer	Art. 9 — risk management
Regulated record written	TPM 2.0 sealed audit entry created	Art. 12 — event logging
Trust score drops < 0.50	Firehorse Snap triggered. Software kill chain: 0.006ms (measured, Helsinki Veriton GN100) VMM hardware kill: <125ms (Firecracker production target)	Art. 14 — human oversight

MCP Integration Path

DWS IQ CLI registers as a native MCP server inside Salesforce Headless 360. All 20 compliance agents are published as Salesforce-callable MCP tools. Agentforce orchestrates them. KYA governs them. The result surfaces in the CRM record — never leaving the Salesforce data perimeter.

Use Case — CBAM/CSRD in Salesforce CRM: A Nordic industrial sales rep opens an Account record. Agentforce calls dws/compliance/run via MCP. The DWS CBAM agent calculates carbon border adjustment liability (EU CBAM Reg. 2023/956) and the CSRD agent maps scope-3 emissions against ESRS E1. KYA governs each handoff. A Firehorse Signed Report appears in Salesforce Files within 60 seconds. The sales rep shares it with procurement — the deal moves forward.

Partner Brief: KYA Connector Brief — Salesforce Headless 360 & Agentforce ISV (CONFIDENTIAL — available on request)
Arctic 15: DWS IQ presenting in Helsinki, June 11–12, 2026. Contact risto@onelifetime.world to arrange a meeting.

11. Contact

Lifetime Oy · Helsinki, Finland
risto@onelifetime.world
dws10.com · onelifetime.world

Partner Whitepaper: Securing the “Shanghai Model” of Agentic Delegation
Back to Legal Hub: View all legal documents & compliance policies

The KYA Standard

Know Your Agent (KYA) v1.1

1. Executive Summary

2. The Three Pillars of KYA

Identity Attribution — The Anchor

Capability Gating — The Leash

Forensic Observability — The Firehorse Audit Trail

3. Technical Architecture: Hardware-Level Isolation

Firecracker MicroVMs

Hypervisor-Level Egress Control

4. Agent Trust Score (ATS)

5. Use Case: The Multi-Agent Control Room

16 Industrial Models

6. BYO Agent Registration Flow

7. Compliance Alignment

8. Subagent Registry v1.1

Capability Manifest

KYA Violation: Unauthorized Path Access

Manifest Schema

Fault Attribution (Liability Mapping)

9. Identity Provider Integration

10. Agentforce Integration — Salesforce Headless 360

MCP Integration Path

11. Contact