Anthropic's Claude Deploys 90+ Legal AI Workflow Agents

Anthropic launched Claude for Legal, releasing over 90 named AI agents on GitHub to run specific legal workflows for law firms, in-house teams, and law schools.

The launch moves legal AI past the question-and-answer stage. These agents do not just retrieve information. They execute recurring tasks end-to-end, under lawyer supervision, covering commercial contracts, corporate transactions, employment matters, litigation, and law school work.

How Agents Work

Each agent carries a job-style name that matches the task it runs. The Vendor Agreement Reviewer checks a vendor MSA against a playbook and produces a redline memo. The NDA Triager assigns a GREEN, YELLOW, or RED priority to inbound NDAs so only the difficult ones reach a lawyer's desk.

Agents run in two modes. Some trigger on a single command. Others run on a schedule, continuously monitoring incoming documents, emails, or contract registers. The Renewal Watcher scans a contract register for cancel-by and renewal deadlines. The Deal Debrief sweeps signed agreements weekly, flags playbook deviations, and prompts the attorney to log context while memory is fresh.

The suite covers six plugin groups: commercial-legal, corporate-legal, employment-legal, litigation-legal, legal-clinic, and law-student. Agents 65 through 68 handle litigation intake, matter briefing, portfolio status, and outside counsel status updates. Agents 78 through 87 target law students, covering bar prep, IRAC grading, case briefing, cold call preparation, and exam forecasting.

Built-In Safeguards

Mark Pike, associate general counsel at Anthropic and the lawyer leading the Claude for Legal rollout, described the model's approach directly: "Opus 4.8 continues our work to make Claude more helpful, harmless, and honest. You can feel it in practice as the model flags uncertainty instead of papering over it."

On the tooling side, Pike stated: "The Claude for Legal plugins are built the same way: source attribution on citations, jurisdiction established during an onboarding interview, and explicit gates before anything is filed, sent, or relied on. The lawyer reviews and verifies; the tooling is designed to make that review easier, never to skip it. Honest model, honest tooling."

Every citation carries a source attribution. Jurisdiction is set at the start through an onboarding interview so agents apply the correct legal framework from the first task. No output moves forward to filing, sending, or reliance without a lawyer clearing an explicit gate.

Customisation and Limits

Each of the 90-plus agents is modifiable in plain language. A lawyer can adjust the underlying skill, the practice profile, and the connectors without writing code. Some technical knowledge remains useful when connecting the agents to an enterprise tech stack.

The suite is published open source under the Apache 2.0 license. Firms can fork the agents, add jurisdiction-specific materials, and retune them to local playbooks. That open architecture matters for teams outside the US common-law practice, where the default settings do not reflect local rules or regulatory frameworks.

The platform requires exclusive use of Claude. Other legal AI platforms, including Harvey, Legora, LexisNexis, and Thomson Reuters, allow firms to select from multiple AI models or mix models across different task types. Claude for Legal does not offer that flexibility.

Where the Value Is

The 12 core plugins and the MCP connectors to existing legal tech tools drew most of the attention at launch. The agents are the more granular layer, and granularity is where practical value concentrates.

A generic contract review tool covers every contract and satisfies no specific need well. An agent scoped to one workflow type, running on a continuous schedule, and tuned to a firm's own playbook addresses a real, recurring task. With over 90 agents across six practice areas, the suite covers a wide range of those recurring tasks out of the box.

The agents can also be refined over time. As workflows change, the underlying skills and connectors update to match. That continuous adjustment keeps the agents useful beyond the initial deployment.

The law student agents add a separate dimension. The Cold Call Prep agent predicts a professor's questions and drills students before class. The Exam Forecaster analyzes past exams from the same professor and identifies likely emphases. The Bar Prep Coach targets MBE and essay practice by subject area. These tools serve a user group that legal AI has largely not addressed until now.

About Anthropic

Anthropic is an AI safety company founded in 2021 and headquartered in San Francisco. It develops the Claude family of large language models. Claude for Legal is Anthropic's dedicated product suite for the legal sector, delivered through 12 plugins, MCP connectors to legal tech tools, and a library of over 90 named workflow agents listed on its GitHub page.