What Is Anthropic Cowork? The 2026 Legal Plugin Explained

What is Anthropic Cowork? It's Anthropic's agentic AI platform: the layer where Claude does multi-step work across tools, not just single-turn chat. The legal plugin Anthropic released for Cowork in February 2026 is open source, free, and configurable. Per the official Anthropic legal plugin page and the knowledge-work-plugins/legal GitHub repo, the plugin ships two commands: /review-contract and /triage-nda; that any law firm or in-house legal department can install on their Claude account and configure against their negotiation playbook. The launch produced a $285 billion combined market cap drop across Thomson Reuters, RELX, and Wolters Kluwer in a single trading session per Canadian Lawyer's reporting. This is the explainer: what it is, what it does, what it costs, and where it fits.

What Anthropic Cowork is: the platform layer

Anthropic Cowork is the agentic AI surface for Claude. Where standard Claude chat handles single-turn or short-context tasks, Cowork is where Claude operates across multiple tools and surfaces in extended workflows. Per Anthropic's product overview, Cowork is the deployment surface for plugins and extended-context workflows.

The platform's structural significance:

• Multi-step task execution. Claude on Cowork can read a document, search for related materials, draft analysis, integrate output back into the source workflow.
• Plugin architecture. Cowork supports installable plugins that add domain-specific commands and behaviors. The legal plugin is one of several available.
• Configurable behavior per organization. Plugins read organization-specific configuration (negotiation playbook, risk tolerance, custom workflows) and adapt their behavior accordingly.
• Open source extensibility. The legal plugin's source is on GitHub. Firms can fork, modify, deploy internally, or build adjacent plugins.

The second-order read: Cowork is where Anthropic competes architecturally against vertical legal AI vendors (Harvey, CoCounsel, Spellbook). The vertical vendors ship pre-built workflows; Cowork ships the platform plus open-source workflows. The competitive frame is platform vs vertical vendor, not foundation model vs vertical vendor.

The third-order read: Cowork's open-source plugin model produces a different innovation pace than the vertical vendor model. Per the 20,000-lawyer workshop analysis, Anthropic ships demos to lawyers directly and lets bottom-up adoption pressure drive enterprise procurement, instead of running enterprise sales cycles.

What the legal plugin actually does: two commands explained

The Cowork legal plugin shipped two commands in its February 2026 release:

One concrete workflow is shown in NDA triage with the Cowork legal plugin.

/review-contract. Runs clause-by-clause review of an uploaded contract against the firm's configured negotiation playbook. Output:

• GREEN flags for clauses matching firm-standard positions.
• YELLOW flags for clauses requiring counsel review (deviations from playbook within negotiable range).
• RED flags for clauses materially deviating from firm-standard positions or regulatory requirements.
• Inline redline suggestions showing recommended modifications with reasoning.
• Counsel handoff summary when the contract requires senior review: highlights the deviations needing decision.

/triage-nda. Rapid pre-screening of incoming NDAs into three tiers:

• Standard approval. NDA matches firm-standard template within tolerable variance: auto-approval recommended.
• Counsel review. NDA deviates from firm-standard positions in negotiable ways: counsel reviews specific flagged clauses.
• Full review. NDA contains material deviations or unusual provisions: full counsel attention required.

Both commands run against organization-specific configuration. The configuration surface (negotiation playbook, risk tolerance, NDA triage tiers) is what determines whether the plugin produces useful output. Configuration time: 8-16 hours of legal-ops time plus 4-8 hours of senior counsel review of the playbook. The Cowork /review-contract clause-by-clause spoke covers the deeper configuration patterns.

What it costs: the plugin is free, the model layer isn't

The plugin itself is free and open source per the GitHub repo. The cost is the underlying Claude subscription. Per Anthropic's pricing:

• Claude Pro at $17/user/month annual or $20/user/month monthly: consumer tier, no Team admin controls. Plugin runs but matter work shouldn't run here per US v. Heppner privilege analysis.
• Claude Team Standard at $20/seat/month annual or $25/seat/month monthly for 5-150 seats: admin controls plus explicit no-training commitment on team inputs. Floor for firm deployments.
• Claude Team Premium at $100/seat/month annual or $125/seat/month monthly: premium seat tier with higher usage caps.
• Claude Enterprise at $20/seat/month + usage at API rates with custom terms, for firms negotiating advanced security, compliance, and data residency.

For a 50-person in-house legal department on Team Standard annual: $12,000/year all-in. For a 200-lawyer mid-market firm on a tier mix: roughly $90,000-108,000/year. For an AmLaw 100 firm on Enterprise tier: custom. The Cowork pricing breakdown covers the per-tier deployment math. The Cowork vs ChatGPT Enterprise comparison covers the head-to-head pricing math against OpenAI's enterprise offering.

How to use the plugin: installation and configuration walkthrough

The deployment path:

• Step 1: Install Claude. Claude Pro for individual lawyer use; Claude Team for firm deployments. Per Anthropic's pricing, most firm deployments start on Team Standard at $20/seat/month annual.
• Step 2: Install the legal plugin. From the knowledge-work-plugins/legal GitHub repo. The plugin source is published openly; install instructions are in the README. Cowork environment supports plugin installation at the user or admin level.
• Step 3: Configure the negotiation playbook. Your firm's standard positions on the clauses you negotiate most: limitation of liability, indemnification, IP ownership, governing law, dispute resolution, data protection, term and termination. The plugin reads the playbook to set GREEN/YELLOW/RED flags during /review-contract.
• Step 4: Configure NDA triage tiers. What gets standard approval, what needs counsel review, what needs full review. Configuration usually starts conservative (most NDAs go to counsel review) and tightens after 30 days of observed accuracy.
• Step 5: Test against sample contracts. Run /review-contract on 10-20 representative contracts to validate the playbook configuration. Adjust based on output quality.
• Step 6: Deploy to lawyers. Train counsel and associates on the commands. Per the Anthropic 20,000-lawyer workshop analysis, Anthropic provides webinar training; firms supplement with internal training tied to their specific configuration.
• Step 7: Instrument metrics. Capture baseline workflow time before activation; track post-activation cycle time, counsel touch rates on triaged NDAs, accuracy of GREEN flag determinations.

The second-order read: configuration time is 8-24 hours of senior legal-ops time depending on practice complexity. Most firms underweight this; the configuration is what determines whether the plugin produces useful output. The in-house deployment checklist goes deeper on the configuration patterns.

Is it safe? the privilege and confidentiality questions

The privilege questions about Cowork legal plugin use are the same as the privilege questions about Claude generally. Per US v. Heppner (SDNY, February 17, 2026), consumer Claude communications were not protected by attorney-client privilege or work-product doctrine. The Heppner ruling addressed consumer Claude specifically.

What that means for Cowork plugin use:

• Use Team or higher tier for matter work. Per Anthropic's terms, Team, Enterprise, and API inputs are not used for model training. Pro inputs may be. Matter work needs Team minimum.
• Direction-of-use matters. Per the Heppner work-product analysis, AI-generated material at attorney direction may carry work-product protection; material generated independently doesn't. The Heppner enterprise privilege defense stack covers the architecture.
• Configuration files contain firm IP. The playbook configuration captures firm-specific negotiation positions. Storage, access control, and confidentiality protection apply.
• Open source means auditable. The plugin source is on GitHub. Firm IT can inspect what the plugin does and how it handles data. Closed-source vendor tools don't offer the same transparency.

The second-order read: Cowork plugin running on Claude Team or Enterprise carries different privilege facts than the consumer Claude facts in Heppner. The privilege analysis depends on the deployment surface plus direction-of-use plus documentation, not on the plugin itself.

The third-order read: per the firm AI policy template, the policy needs to name approved deployment surfaces, citation verification protocol, scratchpad file storage, and AI committee oversight, same architecture whether the firm uses Cowork plugin, Spellbook, Microsoft Copilot, or any other AI tooling. The plugin doesn't change the policy framework.

Where Cowork legal plugin fits: and where it doesn't

Cowork legal plugin is the right deployment surface for several specific scenarios:

• In-house legal departments. Lower budget, configurable workflows, rapid deployment. Per the in-house deployment checklist, 50-person in-house teams land at $12,000-15,000/year on Claude Team Standard plus the free plugin.
• Mid-market firms with significant transactional volume. Configurable NDA triage and contract review at the model layer. Per the procurement checklist for mid-market firms, the standard combination for 50-500 lawyer firms.
• Solos and small firms experimenting with AI workflows. Plugin is free; Claude Pro at $17-20/user/month is on a personal credit card budget. No procurement involvement needed.
• Firms wanting open-source auditability. Closed-source vendor tools require trust; the open-source Cowork plugin can be inspected.

Where Cowork plugin isn't the right answer:

• Firms wanting Westlaw or Lexis content embedded in workflow. Per the LexisNexis-Doctrine acquisition analysis, proprietary content remains a structural moat. Cowork doesn't have research content. Firms with significant litigation or regulatory work need a content-bearing tool layered on top.
• Firms with very high contract throughput in narrow practice specialties. Spellbook's vertical depth plus pre-built clause library may save more configuration time than Cowork's flexibility provides value. The Cowork vs Microsoft Copilot vs Spellbook vendor war analysis covers the comparison.
• Firms running Google Workspace primary or AWS-native. Cowork via Microsoft Foundry doesn't apply; Vertex AI on Google Cloud or AWS Bedrock are the right deployment surfaces.

The bottom line. My take: The Anthropic Cowork legal plugin is the structurally cheapest legal AI deployment for in-house departments, mid-market firms with significant transactional volume, and solos experimenting with AI workflows. Plugin is free and open source; the cost is the Claude subscription (Pro is $17/month on annual billing or $20 monthly; Team Standard is $20/seat/month annual or $25 monthly). The two commands (/review-contract and /triage-nda) replace basic-tier vendor tooling at a fraction of the price for configurable workflows. The plugin doesn't replace proprietary content (Westlaw, Lexis, Practical Law) or vertical specialization (Spellbook). For firms running Microsoft 365 already, the right combination is Cowork plugin plus Microsoft Copilot integration. For BigLaw, Cowork is one of several deployment surfaces: Foundry, Bedrock, or Vertex deployment of Claude with the plugin layered on top is the typical pattern. The plugin doesn't end legal AI vendor competition; it shifts where pricing power sits.

Price sources, checked 22 October 2026: Claude (Anthropic), Claude API token rates. Each figure above is a list price from the vendor's own pricing page, or arithmetic on one. Plans, seat minimums and promotions change; confirm on the vendor's page before you buy.