Reviews

Architecture critiques, trust-boundary reviews, and assessments of how system claims hold up under scrutiny.

April 16, 2026

How AI Agent Systems Treat Governance Failures as Implementation Details

When an AI agent violates scope, misses a policy gate, or produces no receipt, most platforms treat it as a bug to fix — not a designed failure path. The missing recovery architecture is the governance gap.

April 16, 2026

How Zero-Trust Marketing Obscures Trust Boundaries

Zero-trust as a marketing label presents the right vocabulary while obscuring where trust boundaries actually sit. The label does not prove the architecture.

April 16, 2026

Why Third-Party Verifiers Still Fail When the Evidence Path Is Controlled

A third-party verifier that reads only what the system under review provides is not an independent check. Independence of the verifier is necessary but not sufficient — the evidence path must also be independent.

April 13, 2026

Why Most AI Workflow Demos Are Hard to Trust

AI workflow demos routinely conflate capability with governance. This review examines the trust-boundary problems that make most demos unreliable as evidence of production readiness.

April 8, 2026

How RBAC Fails in Multi-Tenant AI Platforms

Standard role-based access control assumes static role boundaries. AI agent platforms break those assumptions when agents act across tenant contexts, escalate through tool-calling, or inherit ambient permissions.

April 7, 2026

Why Compliance Dashboards Are Not Security Evidence

Compliance dashboards present internal state as if it were independently verifiable. They are presentation, not proof. Trace the evidence chain and it ends inside the system.

April 6, 2026

What Breaks When Agents Call External APIs

The trust boundary between an AI agent and an external API is wider than most architectures acknowledge. Failure modes include scope leakage, credential inheritance, response manipulation, and unverifiable execution.