Building in public

Verifiable context transfer for AI agent systems

When AI agents from different vendors collaborate in regulated environments, there is no standard way to verify where context came from, whether it was altered, or what the recipient is allowed to do with it. We are building the protocol layer that solves this.

The Problem

Identity tells you who the agent is. Authorization tells you what it can access. Neither tells you what happened to the data after it left.

Existing standards (OAuth, SPIFFE, MCP, A2A) handle agent identity and access control. But once Agent A has processed data and handed its output to Agent B from a different vendor, the identity and authorization controls that governed Agent A's access do not travel with the transferred context.

Healthcare

Traceability breaks at the vendor boundary

A triage agent hands patient context to a diagnostic agent from a different vendor. The diagnostic agent cannot prove to an auditor that the context it acted on was the same context the triage agent produced.

Finance

Full chain reconstruction is impossible

Market data flows through analysis, risk assessment, and recommendation agents across vendors. FINRA requires full chain reconstruction, but no vendor has a complete picture and none of their logs are cryptographically linked.

Enterprise

Authority doesn't follow the data

When Agent A shares context with Agent B, there is no standard way to restrict Agent B from forwarding raw content to Agent C. Content-level authorization does not exist in current agent protocols.

Compliance

Regulations require what doesn't exist yet

EU AI Act Articles 12 and 25 (effective August 2026) mandate tamper-evident logging and value chain traceability. Multi-vendor agent chains cannot satisfy these requirements without standardized provenance metadata.

What We Build

A protocol for tamper-evident, provenance-tracked context transfer between AI agents

The Context State Attestation Envelope (CSAE) is a structured protocol for transferring AI agent operational context across trust boundaries with cryptographic integrity, per-item provenance tracking, authority controls, and graceful degradation under context window constraints.

Architecture

Four layers. Start simple. Scale to regulated.

Every successful protocol is trivially simple at the base and progressively complex for advanced use cases. CSAE is no different.

Layer 0

Attestation

Content + provenance hash + signature. The irreducible core.

Open Source
Layer 1

Provenance

Full DAG with source types, transformations, per-node hashes.

Open Source
Layer 2

Metadata & Authority

Transformation tracking. Per-item permissions with attenuation.

Commercial
Layer 3

Full Envelope

Six coupled components. Integrity seal. Degradation policies.

Commercial
Standards Landscape

Where the gap sits

The identity and authorization layers are well-served by existing standards. The data flow tracking layer has no standardized solution.

CapabilityOAuth / OIDCMCPA2ASPIFFEIETF DraftsStatus
Agent identityYesVia OAuthAgent CardYesdraft-klrcAddressed
Access authorizationYesVia OAuthAuth schemesNodraft-klrcAddressed
Delegation of authorityScopesSessionNoNoPartialPartial
Provenance of transferred contentNoNoNoNoNoGap
Per-item authority on contentNoNoNoNoNoGap
Cross-boundary integrityNoNoNoNoNoGap
Non-repudiation across vendorsNoNoNoNoNoGap
Regulatory Context

The compliance deadlines are set

These are enacted laws with specific enforcement dates, not proposed legislation. Multi-vendor agent deployments will need compliance infrastructure by these dates.

EU AI Act
August 2, 2026
Articles 12 and 25 mandate tamper-evident logging and full value chain traceability for high-risk AI systems across organizational boundaries.
HIPAA Updates
May 2026
Natural person attribution requirements for AI-processed health data transferred between organizational systems.
FINRA
In effect
Full chain reconstruction including underlying telemetry for automated financial decisions across vendor boundaries.
About

Traverse Labs

Traverse Labs LLC is a technology company focused on structured communication protocols for autonomous AI agent systems. We address how agents transfer operational context across organizational and vendor boundaries with verifiable integrity, authority controls, and audit-trail preservation.

Our work includes a protocol specification, a reference implementation, and active engagement with standards bodies including NIST's National Cybersecurity Center of Excellence on AI agent identity and authorization standards.

Interested in what we're building?

We're open to technical discussion, standards collaboration, and commercial partnerships.

Get in touch