Can EDR See Claude Cowork on Firm Devices?

The blind spot is on the device

Endpoint detection and response monitoring for AI agents raises a practical concern for law firms: an agent may operate on a device that also holds client information. Your endpoint protection may monitor the laptop without providing visibility into every activity associated with the agent.

That distinction matters. A managed laptop does not automatically mean every process running on it is visible to your security team. If Cowork can access files or perform actions, your firm needs to know whether your EDR can observe that activity, apply policies to it, and alert when it behaves in a way your security team considers risky.

Considering Cowork is a useful prompt to check your setup, not a reason to assume your protection already covers it. EDR products differ, and configuration matters. Ask your IT provider or security lead to verify coverage on the exact devices and software versions your staff use. Until you have a tested answer, limit Cowork’s access to firm devices and client files.

This isn’t an argument against using agents. It’s a reminder that convenience and visibility need to move together, especially when client confidentiality, privilege, and matter obligations are at stake. If you’re already reviewing agent access, our guide to locking down AI agent access at a law firm covers the broader controls to consider.

What to check before allowing access

Start with the actual device and workload. Don’t settle for a general confirmation that the laptop is protected. Ask your provider to show whether the EDR agent can see Cowork-related processes and activity, and whether the relevant events appear in the console your security team uses.

Then work through the controls that matter to your practice:

• Visibility: Can your EDR identify Cowork-related activity and report it, or does it only see activity on the device?
• Alerts: Will your team receive an alert for suspicious activity associated with Cowork, or only for other activity on the device?
• Policy enforcement: Can you block or restrict Cowork if a policy is triggered?
• File access: Can Cowork reach matter folders, shared drives, synced desktops, or client data by default?
• Identity and permissions: Does it run with a user’s ordinary permissions, or can it access more than the task requires?
• Records: Can your firm retain enough activity records to investigate a concern and meet internal requirements?
• Exceptions: Who can approve expanded access, and how quickly are temporary permissions removed?

Ask for a demonstration or a test on a non-production device. Confirm what happens when the agent attempts to access a restricted folder. Confirm that alerts arrive where someone will act on them. Have your provider document any gaps and give you a plan to address them.

If the answer is “we think so,” treat that as an open control, not a verified safeguard. Until it is verified, use a separate device with no client data, disable access to shared matter locations, or hold off on using Cowork for firm work. The right interim measure depends on your risk tolerance and obligations, but doing nothing while assuming coverage is the weakest option.

Where the business risk shows up

A security control doesn’t sit apart from the economics of a law practice. The same access that can make an agent useful can expose information or create review work if the firm has not set clear boundaries.

For a law firm, the issue usually isn’t an abstract debate about technology. It’s whether partners know what staff are using, whether the work is approved, and whether the time saved is greater than the time spent checking outputs and cleaning up exceptions. Shadow usage often makes those questions harder to answer. Our piece on shadow AI agents multiplying in law firms looks at how those tools can appear in normal workflows without a deliberate firm decision.

There is also a direct operational cost. Firms may find that attorneys spend time on document review, intake, and matter administration that never makes it onto an invoice. The amount varies by practice mix and process. When partners can’t see how an agent handles work, they may either miss a useful opportunity or spend unplanned time supervising it.

The potential annual leakage for this law-firm use case will vary. That range isn’t a promise that adopting an agent will recover the money. It gives a firm a way to size the operational problem: unbilled work, slow handoffs, and repeatable tasks that consume expensive staff time. A sensible assessment asks where that work occurs, what data the task touches, and what controls would make automation safe enough to consider.

What an agent workflow looks like end to end

A well-designed workflow begins with a defined job, not broad access to a partner’s desktop. For example, a firm may want to reduce delays in handling new inquiries. The agent should have a narrow role, a limited set of approved inputs, a clear handoff point, and a record of what it did.

An Intake Voice Agent can answer calls after hours, during lunch, or when the front desk is tied up. It gathers the caller’s basic information, asks the approved intake questions, checks for potential conflicts through the firm’s defined process, captures the matter, and offers a consultation slot on the firm’s calendar. It should not promise representation or provide legal advice. When a human needs to decide whether to proceed, the workflow routes the information to that person.

A Matter Triage Agent can review incoming forms and emails, classify the likely practice area, score fit against firm criteria, and send the appropriate partner a short brief. Staff still handle judgment calls, conflict resolution, and sensitive exceptions. The agent’s job is to make the incoming queue easier to review, not to make the firm’s legal decisions.

A Document Review Agent can conduct a first pass over contracts, discovery batches, or matter files. It can flag clauses, summarize positions, and prepare an associate-grade memo for human review. The associate checks the source documents, confirms the citations and interpretation, and decides what belongs in the work product.

These are examples of bounded operational work. They aren’t a reason to give an agent unrestricted access to every matter folder. The implementation should specify which documents the agent can read, whether it can write or move files, what it can send outside the environment, and who reviews the result. Access should be limited to the task, and logs should make the work inspectable.

The connection to Cowork is straightforward. If your staff plan to use it on a firm device, understand what it can access and what EDR can monitor before placing client material within reach. If your firm is considering a separate agent workflow, define the task, permissions, and human checkpoints before choosing the tool.

Put a number on the work, then set the boundary

A partner can start with a small sample instead of a firm-wide program. Pick one workflow, such as new inquiry handling or first-pass document review. Track how many items arrive, how long they wait, how much staff time each takes, and how often a human has to correct or complete the work.

For intake, check how quickly calls and forms receive a response, including evenings and weekends. Some after-hours intake may fail to convert, though the actual rate depends on location, practice area, and how callers are handled. For document review, associate time costs vary depending on seniority and market. A first-pass process may be worth examining, but only if the firm can protect client material and verify the output.

Then ask what minimum access would support the test. Could the agent work with a redacted sample? Can it access only a designated folder? Can it run under a separate identity? What event would make the firm pause the test? These are practical questions for the partner responsible for the process, the person responsible for security, and whoever manages client confidentiality obligations.

If you want to map the workflow, risk, and expected value in one working session, Book a 60-min Omni Audit. We’ll identify where manual work is accumulating, where access needs a tighter boundary, and which process is worth testing first. You can also see Omni for law firms before booking.

A practical checklist for intake

If new-client intake is one of the processes you want to review, use the AI Client Intake Checklist for Law Firms as a working worksheet. It helps you note what happens from the first call or form through conflict checks, qualification, routing, and appointment booking. You can download the checklist here and use it with your intake lead to identify handoffs that need clearer ownership.

The checklist is about intake, not endpoint security. Keep those reviews connected, but don’t confuse them. An intake workflow can be well designed and still be unsafe if an agent has broad access to client files or runs where your EDR can’t see it. For a wider view of oversight, see our guide to AI agent governance frameworks for law firms.

What an Omni Audit produces

An Omni Audit is a working session, not a slide presentation. We map a real process, find the points where time or client opportunities are being lost, and identify where agent access needs limits or human review.

The session can produce a workflow map, a prioritized opportunity list connecting manual work to potential time or revenue impact, and a first-step implementation plan identifying a bounded test, required access controls, and the people who need to approve it.

There is no deck to sit through. The aim is to leave with a practical next step that fits the firm’s operations and security requirements. If the process doesn’t make sense to automate, that’s a useful outcome too.

Make monitoring a gate, not an afterthought

If a firm device contains client information, don’t approve Cowork access based on an assumption that endpoint protection covers all Cowork-related activity. Ask your provider to demonstrate monitoring, test the policy and alert path, and restrict access until you have a verified answer.

At the same time, identify the work your firm wants to improve. Intake delays, document review, and matter administration can consume real capacity. A narrow workflow with limited permissions, human review, and visible records gives you a more responsible place to start than broad access across the firm.

For a focused review of the workflow and the controls around it, Book your 60-min Omni Audit. You can also review the AI audit for law firms and decide which process is worth putting on the table.