As AI assistants get plugged into the software companies actually run on, a quiet question keeps surfacing: who decides what the assistant is allowed to touch? On August 24, Anthropic gave enterprises a clearer answer. It made enterprise-managed authorization for its connector framework generally available, shifting control over Claude's integrations from individual employees to the IT department.
The framework in question is built on the Model Context Protocol, the open standard Anthropic introduced to let AI models talk to outside tools and data in a consistent way. Connectors are the specific bridges into services like a company's project tracker, its documents or its monitoring stack. Until now, wiring one up often meant each employee granting access on their own. The new setup lets an administrator approve, configure and revoke those connections centrally, and set how they authenticate.
Which tools, and why it matters
The generally available release widens the roster of supported services to include Datadog, Notion and Slack, joining earlier integrations for Asana, Atlassian, Canva, Figma, Linear and Supabase. Anthropic says Exa, Miro and Zoom are on the way. On its own that is a routine expansion. The governance layer underneath is the real story.
Once an assistant can read your incident logs, edit your design files and post in your team channels, connector sprawl stops being a convenience and starts being a security surface. Centralised authorization means a security team can see every tool Claude is reaching into, cut off one that should not be there, and apply the same rules across a whole organisation instead of trusting thousands of individual choices.
The move fits a broader pattern across the industry. Standards for how AI systems connect to tools, and to each other, are gradually being pulled into neutral, shared governance, as happened when Google handed its agent-to-agent protocol to an independent foundation. And the commercial stakes rise as those connections start carrying real transactions, the direction pointed to by Alipay's push into agent-driven commerce.
For most workers the change will be invisible. That is rather the point. The interesting part of enterprise AI right now is not what the models can say, but the plumbing that decides what they are permitted to do, and who gets to hold the valve.
Commentarii · 0