Claude Cowork adds enterprise controls

Anthropic is trying to solve the part of workplace artificial intelligence that usually kills a rollout: the security review. On April 9, 2026, it moved Claude Cowork out of research preview and added admin controls that let companies decide who can use what, how much they can spend, and what gets connected. (claude.com) Claude Cowork is the desktop version of Claude that can take actions across work tools instead of just answering questions in a chat box. Anthropic says teams are using it for tasks like drafting deliverables and keeping people updated, which is why the company is now packaging it for wider company deployment. (claude.com) The new layer starts with role-based access controls, which means a sales team and a finance team do not have to get the same powers. Anthropic’s help docs say Enterprise plans can assign custom roles at the group level, so permissions come from the role instead of from one default setting for everyone. (claude.com) (support.claude.com) It also adds group spend limits, which is the budget version of a circuit breaker. Anthropic says admins can cap usage by group, so one heavy-use team does not accidentally run up a company-wide bill while another team is still testing whether the tool is useful. (claude.com) Then there is analytics, which is the part procurement teams ask for after the demo ends. Anthropic says admins can now see adoption and usage data for Claude Cowork, and its help center lists an Enterprise Analytics Application Programming Interface for engagement and adoption data. (claude.com) (support.claude.com) Anthropic also expanded OpenTelemetry support, which is a standard way to trace what software is doing across systems. In plain terms, it gives information technology teams a way to watch Claude Cowork activity with the same observability plumbing they already use for other business software. (claude.com) (support.claude.com) The product side of the launch is a plugin system that turns Claude into a specialist instead of a general helper. Anthropic’s plugin docs say each plugin can bundle skills, connectors, and sub-agents into one package, and the public plugin directory says those plugins work across Claude Code and Cowork. (support.claude.com) (claude.com) That matters because connectors are where enterprise deals usually get tense. Anthropic’s docs say Cowork can connect to services like Google Drive, Gmail, Slack, and DocuSign, but those connections reach external services through Anthropic’s cloud, which is exactly the kind of detail security teams want spelled out before approving a rollout. (support.claude.com) Anthropic is also building a marketplace dynamic around those plugins instead of keeping everything first-party. Its public GitHub repository already lists role-specific plugin packages for fields like finance, legal, sales, marketing, and human resources, plus partner-built entries such as Zoom. (github.com) The timing explains the packaging. Anthropic spent the last year adding business-plan controls across Claude, including spend controls, analytics, audit tooling, and compliance features, and Claude Cowork now inherits that same “ready for admin review” posture instead of looking like a clever desktop experiment. (www.anthropic.com) (support.claude.com 1) (support.claude.com 2) So this launch is less about a new chatbot and more about turning an employee-side assistant into software a company can actually buy. The feature list Anthropic shipped on April 9 reads like a checklist from an information security team: permissions, budgets, telemetry, controlled integrations, and a paper trail for who did what. (claude.com)