Research explainer

AIUC-1 Q2 2026: Why agent trust is shifting to MCP, A2A, identity, and third-party monitoring

AIUC-1's Apr 15, 2026 update said 14 requirements and 23 controls were updated or added. The plain-English message is that agent assurance is getting more protocol-aware: approved interfaces, verified identities, logged actions, and monitored third parties now matter more in buyer review.

Protocol-aware agent assurance

This page explains what the Apr 15, 2026 AIUC-1 research update means for buyers and publishers. It does not claim that CraftedTrust itself is AIUC-1 certified. It explains why reviews are moving toward approved tools, verified identities, logged actions, monitored third parties, and exportable technical evidence.

What changed in plain English

AIUC-1's Q2 2026 update pushed agent assurance closer to the protocols and runtime behaviors buyers actually have to approve.

Approved interfaces

Approved MCP servers and runtime containment

The update says agent connections should stay on approved MCP servers and that MCP execution environments should be contained at runtime. Control examples: B006.1 and B006.3.

Stronger auth

Authentication and encrypted transport across AI interfaces

The update extends caller authentication and encrypted transport expectations across model APIs, MCP, and A2A. Control examples: B008.2 and B008.3.

Tool governance

Tool calls need authorization, validation, and logging

Tool-call authorization and input or output validation now apply more directly to MCP tool actions, with logging expectations expanded to include MCP tool metadata. Control examples: D003.1 and D003.3.

Third-party risk

Third-party access monitoring is now front and center

AIUC-1 made third-party access monitoring mandatory and kept vendor due diligence in scope, reflecting how MCP servers and third-party agents can appear dynamically at runtime. Requirement example: E009.

Identity

Agents need verifiable identities and scoped permissions

The update adds cryptographically verifiable agent identities and permission-ready access architecture, including just-in-time style controls, so access can stay narrow and reviewable. Control examples: A003.3 and A003.4.

Why MCP matters now

MCP is no longer just a developer integration detail

The current MCP authorization spec says HTTP-based transports should use OAuth 2.1, authorization server metadata, PKCE, secure token handling, and HTTPS. That turns MCP review into a real identity, authorization, and transport question instead of a simple feature checklist.

Why A2A matters now

A2A expands the buyer surface from tools to agents talking to agents

Google introduced A2A as an open protocol that complements MCP for multi-agent interoperability. Once agent-to-agent coordination becomes part of the architecture, buyers have to review not just tools and prompts, but also agent identity, delegation, transport, and logging between connected agents.

Why this is now a buyer and procurement issue

Third-party agent and MCP access is becoming a live approval question because the attack surface changes with each new connection.

Runtime discovery

Connected agents and MCP servers can appear dynamically

That means approval cannot stop at a static vendor questionnaire. Buyers need to know what is actually allowed at runtime.

Review depth

Checklist-only governance is not enough

Policies still matter, but buyers increasingly ask for technical proof around interfaces, identities, tool-call controls, transport, and monitoring.

Evidence speed

Teams need evidence they can export quickly

Security review, procurement, and platform teams need buyer-ready evidence for MCP, A2A, and connected agents without rebuilding the packet each time.

How CraftedTrust maps to the problem

CraftedTrust is the evidence layer that helps teams show both operational governance and technical proof.

Registry / MCP Trust

Approved interfaces, scans, and public trust

MCP Trust gives buyers a public registry, scan depth, certification state, and linked research so approved interfaces are easier to compare and explain.

Identity

Verified organizations, roles, keys, and access boundaries

CraftedTrust Identity supports the identity layer around organizations, roles, keys, and access boundaries that buyers increasingly ask to see.

Governance workspace

Evidence, findings, reports, and monitoring

AI Governance and buyer-diligence flows organize findings, evidence gaps, reports, and monitoring notes into a review-ready record.

Touchstone

Policy and regulatory updates that explain why the controls matter

Touchstone research tracks standards, advisories, and regulatory movement so teams can connect technical findings back to buyer and governance expectations.

Runtime / trace direction

Approvals, guardrails, and runtime-proof direction

Runtime Gateway and approval-trace work are the direction for cases where point-in-time review is not enough and buyers need stronger runtime proof, guardrails, or action history.

More technical than checklist-only governance, more useful than a public score alone

That is where protocol-aware agent assurance is heading. Public scores still help. Governance programs still matter. But buyers increasingly need both operational governance and technical proof across MCP, A2A, identities, tool actions, and third-party monitoring.
Source note This explainer is grounded in the Apr 15, 2026 AIUC-1 Q2 update, AIUC-1's ISO 42001 deep dive, the current MCP authorization specification, and Google's A2A protocol announcement.