MCP3 min
Why MCP OAuth does not settle tool authorization
The MCP authorization specification protects HTTP resources; enterprises still need decisions over individual tool actions and arguments.
Research for systems that act.
Analysis of autonomous authority, agent security, runtime authorization, MCP, machine identity, standards and high-assurance systems. Read the sources, follow the architectural consequence, and test the action boundary.
Featured analysis / Standards
What NIST’s agent identity work means for runtime authorization
NIST’s agent identity effort makes attribution a foundation. The action boundary still needs a task-specific permission decision.
Read the analysis ↗Latest research
New analysis
Published work is dated and source-led. The reference library holds the durable technical explanation.
Standards3 min
OWASP Agent Control Standard: the first controls to implement
A practical reading of OWASP’s new agent-control material through enforceable action boundaries and evidence.
MCP3 min
MCP security guidance: where authentication ends and authority begins
NSA and the MCP specification illuminate a credential and server trust boundary that tool-level permissions must complement.
Agent Identity2 min
Why agent identity alone cannot control consequential actions
Identity improves attribution, while task authority and execution checks govern what a verified agent may cause.
Runtime Authorization2 min
What happens when a parent agent’s authority is revoked?
Child grants, cached decisions and paused tasks need a defined invalidation path when an ancestor is cancelled.
Security Research2 min
The confused deputy problem in multi-agent systems
A privileged parent or tool can be induced to spend its authority on a child request whose origin lacks permission.
Architecture2 min
Tool discovery is not tool authority
A visible MCP tool catalog helps agents plan; it does not grant permission for every invocation or argument.
September 2026 analysis
New source-led work
NIST’s AI Agent Standards Initiative: the interfaces still to define
The February 2026 initiative seeks secure agent interoperability. An architecture review can already test identity, delegated permission and effect evidence as separate interfaces.
An IETF WIMSE agent-delegation draft is a proposal, not a deployment rule
A July 2026 Internet-Draft composes OAuth mechanisms for agent delegation. Its useful lesson is to test attenuated credentials and revocation while keeping draft status explicit.
Editorial taxonomy
Seven lenses, one operating question.
Which permission, identity or control actually governs the action? These topic labels organize the publication; they are not thin indexable archive pages.
StandardsSecurity ResearchArchitectureMCPAgent IdentityRuntime AuthorizationMarket Analysis
Explore the authority library ↗