Claude Cowork gets to the mobile: productivity from the phone and new security risks

Anthropic is testing the arrival of Claude Cowork a mobile, an interesting extension of its agentic desktop-oriented mode that allows you to run and monitor long tasks of the assistant directly from the phone. If you don't know Cowork, it works as a layer that brings continuous running capabilities - previously seen in Claude Code for developers - to the work of documentation, file handling and day-to-day reporting; you can see the filtered catches in X and Claude's general description on Anthropic's website in https: / / www.anthropic.com / claude.

The images provided suggest that the mobile experience will act above all as a remote control for the tasks that run on your computer: start, direct and review progress from your phone, while heavy work continues to run on the PC and continues in the background even if you close the application. This retains the Cowork architecture, in which the agent access the files you decide to share and operates locally on the host team.

In practical terms, this approach can be very useful: long-term tasks, analysis of directories to clean space, compilations that take hours or generation of automated documents and spreadsheets are cases in which telecontrol from mobile facilitates supervision without having to be left next to the computer. In similar tests, users have detected that Cowork explored local partitions and found files that consumed much of the storage, helping solve compilation failures and space bottle necks.

However, moving control to a mobile device changes the risk profile. Running an agent with access to local files and persistence in the background increases the attack surface from endpoint: a committed mobile could send control commands; a vulnerability in the app or in the communication layer could allow unauthorized orders; and the fact that the work continues after closing the app raises questions about credentials, persistent tokens and traceability. In addition, local execution reduces certain cloud risks but makes the job the critical point to protect.

To mitigate these risks I recommend concrete measures: limit Cowork's access to only necessary folders and avoid sharing full units; run Cowork on isolated environments (virtual machines or containers) when handling sensitive data; apply access controls and strong authentication for mobile connection (MFA, session limits and tokens); monitor telemetry and outgoing traffic with EDR / IDS and record the actions of the audit agent. Security operations should add Cowork to the list of managed assets and create specific rules in ICES / EDR to detect atypical behaviors.

For safety equipment and product managers it is key to require safety guarantees by design: encrypted and authenticated channels, signature of binaries, least privileged granules policies, attestation mechanisms that link the mobile session to host equipment and unchanging activity records. It is also recommended that integration be subject to break and attack simulation to validate that controls detect and block abuse - a good practical starting point is in the technical whitepaper of Picus on attack simulation Here. which helps to check detection rules in ICES and EDR.

In short, the arrival of Cowork on the mobile promises to improve productivity by allowing long tasks to be controlled from anywhere, but it carries new risks that should be managed before it is adopted on a scale. The practical recommendation is to pilot the function in controlled environments, update security policies and endpoint, and demand that suppliers clear about permits, persistence of credentials and audit before they deploy it in production.