AI Agents at Law Firms: OpenClaw, Claude Duet, ChatGPT Dots and Grok Bot
A new kind of AI tool spread through 2026: the always-on agent. Instead of answering a question and stopping, it runs on its own computer, signs in to email, calendars and other apps, keeps working when you step away, and comes to you with what needs attention. OpenClaw, free and open-source, came first. Grok Bot, ChatGPT dots and Claude Duet followed from xAI, OpenAI and Anthropic, each running in the cloud. For a law firm, the question is which client information an agent can reach, on which account, and who approves what it does.
Its own computer. The agent works on a computer in the cloud, with its own browser, so your laptop does not need to be on.
Your apps. It signs in to the accounts you connect: email, calendar, documents, messaging and business software.
Background work. You hand it a task and it keeps going, then reports back when it is done or needs a decision.
It comes to you. It watches your connected apps and messages you when something needs attention.
Approval before acting. Each vendor sets rules for which actions the agent may take alone and which wait for you.
A chat assistant answers what you ask. An agent acts in your accounts, which is why the account it runs on and the apps it can reach matter more.
How do the four compare?
As of November 2026. These products are new and change often.
Agent
Who makes it
Plans
How it works
Before it acts
OpenClaw
Free, open-source software, formerly called Clawdbot and Moltbot
Free. The firm or user runs it on its own computer, virtual machine or container
Downloads and runs skills, which are code, from sources including ClawHub, its public skills registry. Microsoft describes it as having limited built-in security controls
Depends on how it is set up
Grok Bot
xAI (now publishing as SpaceXSI)
Included with SuperGrok, SuperGrok Plus and SuperGrok Heavy, and with Cursor Pro, Pro+, Ultra and Teams plans. Team Bots, shared by a team, are on Team and Enterprise plans
Each Bot runs on a computer in the cloud with browser and terminal access. A Bot has no access by default and reaches only the accounts you sign it into. Works in Slack, Microsoft Teams and video calls
xAI states Bots ask before taking sensitive actions and respect each person's permissions
ChatGPT dots
OpenAI
Pro and Business Premium in eligible markets. Enterprise, Edu and Healthcare workspaces as a beta that an admin turns on
Each dot has its own cloud computer and browser, connects to apps through ChatGPT's plugins, and can be messaged in ChatGPT, Slack and Teams. Background research uses read-only tools
Built-in rules decide when to ask. Custom Rules let you allow, require approval for or block specific actions. Some tasks, such as changing a password, always stay with you
Claude Duet
Anthropic
A research preview by waitlist, for Pro, Max and Ultra plans in the US
A separate app with its own virtual machine and browser. Sends a morning brief, checks connected accounts every few hours, and uses Claude's connectors and plugins
Shows exactly what it is about to do and waits for approval before it sends a message, makes a payment, signs a document or deletes anything. Trading securities and moving money between accounts are off-limits even with approval
As of November 2026. These products are new and change often.
What do agents mean for client confidentiality?
The account sets the data terms. OpenAI does not use content from ChatGPT Business, Enterprise or Edu workspaces to improve its models by default; on personal plans the user controls it. Anthropic does not train on content from its Team and Enterprise plans by default; on its individual plans, such as Free, Pro and Max, the user chooses in privacy settings whether chats help improve Claude. Claude Duet's preview runs on individual plans (Pro, Max and Ultra), so check that setting before any client matter goes near it.
An agent sees what it is signed into. Connect the fewest accounts that do the job. An agent signed into the firm's email can read every client message in it.
Shared agents carry context. Grok's Team Bots learn from everyone they work with, and dots carry context across every channel you use. A firm with ethical walls should check how sharing is scoped before a shared agent goes near matter work.
Approval rules are settings, not guarantees. Set which actions always wait for a person, such as anything sent to a client, a court or opposing counsel.
Ethics duties still apply. ABA Formal Opinion 512 applies existing duties to generative AI: understand the tool's benefits and risks, evaluate what happens to client information, consider whether client consent is needed, supervise the output, and do not bill for time the tool saved as though a person spent it.
Is OpenClaw safe for a law firm?
Microsoft's security team published guidance on OpenClaw in February 2026. It says OpenClaw should be treated as untrusted code execution with persistent credentials, and that it is not appropriate to run on a standard personal or enterprise workstation. Where an organization decides to evaluate it, Microsoft says to run it only in a fully isolated environment, such as a dedicated virtual machine or a separate device not used for daily work, with dedicated credentials and no sensitive data. For most environments, Microsoft says, the right decision may be not to deploy it.
The risk comes from what OpenClaw is built to do: read text from anywhere, install skills that are code, and act with the credentials it holds. Microsoft notes public reports of malicious skills appearing in public registries.
For a law firm, that means OpenClaw does not belong on a computer that holds client files or signs in to firm accounts. Because it is free and installs in minutes, it is worth checking whether anyone at the firm has already installed it.
What should a firm's AI agent policy say?
Which agents are approved, and on which kind of account: a firm workspace, not a personal plan.
Which apps an agent may connect to, and which stay off-limits, such as the trust account and the document management system.
Which actions always need a person: sending to clients, courts or opposing counsel, filing, paying and deleting.
Who reviews agent work before it leaves the firm.
How to switch an agent off and disconnect its accounts quickly.
No self-installed agents, such as OpenClaw, on firm computers or with firm accounts.
How does Big Mode Consulting help firms with AI agents?
Inventory. We find which AI tools and agents are already in use at the firm, and on which accounts.
Policy. We write the firm's AI agent policy: approved tools, accounts, connections and approvals.
Setup. We move AI use onto business workspaces and connect agents only to the apps they need. We also build MCP connectors with read and write access for Filevine, Clio and MyCase.
Security. We lock down the computers and accounts agents touch, as part of our cybersecurity work.
Anthropic's proactive agent app, opened as a research preview in October 2026. It comes to you with what needs your attention, completes tasks you hand it, keeps working when you step away, and checks with you before it acts. The preview is by waitlist for Pro, Max and Ultra plans in the US.
What are ChatGPT dots?
OpenAI's always-on agents, announced in September 2026. Each dot has its own cloud computer and browser, connects to apps, and can be messaged in ChatGPT, Slack and Teams. They are rolling out on Pro and Business Premium plans, with an Enterprise beta that an admin turns on.
What is Grok Bot?
xAI's always-on AI teammates. Each Bot runs on a computer in the cloud with browser and terminal access, signs in to the apps you give it, and can work in Slack, Microsoft Teams and video calls. It is included with SuperGrok and several Cursor plans, and Team Bots are on Team and Enterprise plans.
What is OpenClaw?
Free, open-source agent software, formerly called Clawdbot and Moltbot, that a user runs on their own computer. Microsoft's security guidance says it should not run on a standard personal or work computer, and should be tested only in an isolated environment with dedicated credentials.
Can a lawyer use an AI agent with client email?
Only with care. Use a business workspace whose terms keep client content out of model training by default, connect only the accounts the task needs, require approval before anything is sent, and follow ABA Formal Opinion 512's duties on confidentiality and supervision.
Do AI agents train on client data?
It depends on the plan. OpenAI does not use content from Business, Enterprise or Edu workspaces to improve its models by default, and Anthropic does not train on Team and Enterprise content by default. On personal plans of both, the user's privacy settings decide. Check each vendor's terms for the plan you use.
Which AI agent is best for a law firm?
The one the firm can run on a business account, connected only to what it needs, with approvals set before it acts. Claude Duet's preview is on individual plans, ChatGPT dots reach business plans through Business Premium and an Enterprise beta, and Grok Bot offers Team and Enterprise options. OpenClaw needs an isolated environment.