Anthropic has released Claude Cowork, a new feature that allows its AI agents to operate inside desktop applications. The company claims the agents can learn from screen recordings, but those claims have not been independently verified.
How Claude Cowork Works
Claude Cowork gives AI agents the ability to interact with desktop software directly. Instead of working only through a chat interface, the agent can now click buttons, type text, and navigate menus within applications like spreadsheets, email clients, or web browsers. This opens up the possibility of automating complex workflows that span multiple programs. For example, an agent could pull data from a database, format it in a spreadsheet, and then generate a report — all without the user having to switch between apps manually.
The feature is part of Anthropic's broader push to make AI more useful in everyday computing tasks. By embedding the agent inside the desktop environment, the company aims to reduce the friction of moving between AI tools and traditional software.
The Screen Recording Question
Anthropic has said that Claude Cowork's agents can learn from screen recordings. That would mean the AI could observe a user's actions — such as how they fill out a form or organize files — and then replicate those steps on its own. The idea is that the AI could adapt to individual workflows without explicit programming.
But the company has not provided evidence to back up this claim. No third-party tests have been published, and Anthropic has not released technical details about how the screen recording data is handled. Questions remain: Is the recording done locally or sent to the cloud? How long is the data stored? Can users opt out? Without answers, the claim remains unverified.
Privacy and Trust Concerns
The lack of verification is especially significant given the sensitive nature of screen recordings. A user's screen can contain personal emails, financial data, or confidential documents. If the AI is recording that information, users need to know exactly what is being captured and how it is protected.
Anthropic has not addressed these concerns publicly. The company's silence on the matter leaves users and developers in the dark. For businesses that handle sensitive data, the unverified claims could be a dealbreaker. They may hesitate to adopt Claude Cowork until independent audits confirm the safety of the screen recording feature.
Claude Cowork is now available. Whether Anthropic will release more details about the screen recording capability — or submit it for independent review — remains to be seen. For now, the feature's most intriguing promise is also its most uncertain.
