When an AI Work Session Follows You Everywhere, the Risk Surface Follows Too
Claude Cowork’s move to web and mobile turns a desktop-bound assistant into a cross-device workflow, and that shift changes how security teams should think about trust, persistence, and control.
AI tools are no longer staying put on one machine. Claude Cowork is now expanding beyond desktop into web and mobile, with task sessions that can persist across devices and continue in the background without an active connection. That makes the product easier to use, but it also makes the security model more complicated: once a workflow can outlive a single device, it starts to look less like a chat window and more like a managed execution plane.
Fast Facts
- Claude Cowork is being rolled out on web and mobile, not only on desktop.
- Task sessions can persist across devices and keep running in the background.
- The beta starts with Max-tier subscribers over the coming weeks.
- Additional plan tiers are expected to follow after the initial rollout.
- The product is described as an agentic workspace, which places tool use and multi-step work at the center of the design.
What this architecture changes
The important detail is not just convenience. Cross-device persistence usually implies a server-side session model, where task state is held remotely and resumed from different clients. In practical terms, that means the user experience is no longer tied to whether a laptop is open or a phone app is active. The work thread can continue, and that continuity is exactly what makes agentic systems powerful.
From a defensive perspective, that same continuity expands the trust boundary. When an AI agent can keep working after the user has stepped away, the key questions become: what can it access, what can it act on, and how easily can that access be revoked? In agentic systems, those are not abstract questions. They determine whether a harmless request stays bounded or turns into a broader workflow touching files, browser content, or connected tools.
There is also a governance problem hiding inside the convenience. A task started on one device and resumed on another can make oversight harder, especially in workplaces where multiple devices, shared accounts, or delayed approvals are common. The broader risk is not that background execution is inherently unsafe, but that it raises the cost of a mistake. If the wrong prompt, permission, or connector is in play, the agent can continue operating while nobody is watching closely.
For security teams, the practical lesson is simple: treat cross-device AI sessions like privileged automation, not like ordinary messaging. Limit what the agent can reach, review which apps and folders are connected, and assume that persistence increases the blast radius of poor configuration. Plan-gated beta access can reduce exposure during early rollout, but it does not remove the need for least privilege and human oversight on sensitive actions.
At the time of writing, the available information supports a risk analysis, not a claim that broad compromise or misuse has already occurred. The more useful takeaway is that AI workspaces are becoming durable, mobile, and operationally persistent, and security controls have to evolve with them.
Conclusion
Claude Cowork’s expansion is a reminder that the next security boundary is not just a device. It is the session itself. Once an AI workflow can follow a user across platforms and keep running in the background, the real question shifts from where the tool lives to how tightly it is governed. That is where modern AI security will be won or lost.
TECHCROOK
hardware security key: A hardware security key is a simple way to add strong MFA to the accounts that host AI workflows. If a session can move between web and mobile, protecting the underlying login matters more. Use a key for supported services, keep a backup key in a separate place, and pair it with account recovery controls.
WIKICROOK
- Agentic workspace: A system where an AI can carry out multi-step work using tools and context rather than only answering prompts.
- Cross-device session: A work session that can be resumed on different devices without starting over.
- Background execution: Processing that continues even when the user is not actively connected.
- Trust boundary: The limit of what a system is allowed to access or influence without extra controls.
- Least privilege: A security principle that gives a tool only the minimum access it needs to function.
