Assess ▼
Enable ▼
Build ▼
Sustain ▼
Govern ▼
Research ▼
Resources ▼
About ▼
Contact

Can Claude act without an advisor's approval?

By Dr. Leigh Coney, Founder of WorkWise Solutions · Updated October 20, 2026

Short answer: The advisor plugin is designed to ask first: Anthropic says the advisor must explicitly approve any action that writes back to an external system, such as a CRM update or a client draft, and Claude does not place trades or move money. But the plugin's approval rule is enforced by the model following instructions, and Cowork has an organization setting, Allow Automatically approve mode, that is on by default. Turn it off, keep Always allow for connector tools off, and set write and delete tools to Needs approval or Blocked, so the software enforces approval.

Start with what the plugin promises. Its README says: "The advisor must explicitly approve any action that writes back to an external system (e.g. CRM updates, client drafts)." Anthropic's advisor page adds that Claude "doesn't place trades or move money."

Then read the next part of the same README. Some guardrails are structural, such as read-only tools for the agents that parse files. Approval before any write sits in a different group: rules "enforced by the model following them, not by the runtime."

That distinction matters for a fiduciary. A rule the model follows is a strong habit. A rule the software enforces is a control, and an examiner will ask about controls.

Cowork gives you three, all set by an Owner:

  • Allow Automatically approve mode. On by default, so members can let Cowork proceed without asking. Turn it off for client work.
  • Allow Always allow for connector tools. Off by default. It decides whether people can skip per-task approval for write-capable connector tools. Leave it off.
  • Connector tool permissions. For each connector, set the read-only tools and the write or delete tools to Always allow, Needs approval or Blocked. Users can't override the organization's setting.

One catch: read-only tools skip approval only when the connector labels them read-only, and Anthropic says most custom connectors don't. That errs on the safe side, and it means advisors will see more approval prompts than they expect. Tell them why before launch.

Approval thins out in two more places. Computer use, where Claude clicks and types on the screen, runs without the permission checks that gate other Cowork tools. And Anthropic warns against scheduling tasks that send messages on your behalf or take actions that are hard to undo. Keep both out of client work.

Deletion has a hard stop: Cowork asks before permanently deleting any file. Everything else is your configuration, and Anthropic's safety guide puts the responsibility where examiners will: "You remain responsible for all actions taken by Claude performed on your behalf."

Want it set up for you?

I set it up for RIAs. Through WorkWise Solutions, the Claude for Financial Advisors setup covers the plan decision, the connectors, your firm's rules in the skills and archive capture, at a fixed price: $10,000 for firms of 1 to 10 people or $20,000 for 11 to 25, in two to four weeks.

Schedule a Call