In-House Counsel Anthropic Claude Deployment Checklist

In-house legal departments were the structurally happiest readers of Anthropic's February 2026 Cowork legal plugin launch. They've spent five years trying to build internal NDA triage and clause review tooling without engineering budget. Per Anthropic's pricing, a 50-person in-house team running on Claude Team at $20-25/seat/month lands at $12,000-15,000/year total — for the model layer plus the open-source Cowork plugin. That's the floor for workflows that vendor procurement would historically price at $50,000+ annual minimum. This checklist covers what to deploy, in what order, and what to lock down before activating the seats.

Step 1: Decide whether you're replacing existing tooling or layering on top

In-house departments approach Claude deployment from one of two starting positions, and the deployment plan depends entirely on which:

- Layering on top of existing legal-ops tooling (Ironclad, Spellbook, LinkSquares, ContractWorks). Claude becomes the model layer underneath specific workflows where the existing tool isn't deep enough — usually first-pass NDA triage or non-standard contract review. Existing tools stay; Claude fills the gaps. - Replacing existing tooling (typically when current vendor contract is up for renewal or current tool is underutilized). Claude becomes the primary surface; the Cowork legal plugin handles the clause review and triage workflows; the existing vendor either gets replaced or downgraded to lighter usage.

The second-order read: layering is structurally easier and politically cheaper but doesn't capture the procurement-budget shrink that makes Claude attractive in the first place. Replacing is harder and slower but produces the cost reduction that justifies the project to finance.

The third-order read: most successful in-house Claude deployments start as layering, then convert to replacement at the next vendor renewal cycle. The 12-month sequence: deploy Claude alongside existing tooling, document the workflow coverage gap, present the data at renewal time, drop or downgrade the existing vendor.

Step 2: Pick the tier — Team is the floor, never Pro

In-house tier selection follows the same rule as outside firm deployments: Claude Team is the floor; consumer Pro is not appropriate for matter work. The reason is structural — per Anthropic's terms, Team, Enterprise, and API inputs are not used for model training, but consumer Pro inputs may be. Privileged or commercially sensitive in-house work cannot run on consumer Pro.

Tier breakdown for in-house deployments:

- Claude Team Standard at $20/seat/month annual or $25/seat/month monthly for 5-150 seats — admin controls plus no-training commitment. Floor for in-house work. - Claude Team Premium at $100/seat/month annual or $125/seat/month monthly — premium seat tier for power users running multi-day diligence workflows with multi-session memory. - Enterprise at $20/seat/month + usage at API rates with custom terms — for in-house departments that need advanced security, compliance, and custom data residency. Most 50-200 person legal departments don't need this; it matters at 500+ person legal departments at multinational corporations.

For a 50-person in-house team: provision Claude Team Standard for the full team. Annual cost lands at $12,000 (annual billing). For a 200-person legal department: 80% Team Standard, 15% Team Premium for senior counsel running M&A and complex deal work, 5% on dedicated API access for tooling builds. Annual run-rate roughly $90,000-108,000.

Step 3: Configure the Cowork legal plugin against your actual playbook

The Cowork legal plugin is open source and free. The configuration step is what determines whether it produces useful output for your specific in-house workflows.

The configuration surface:

- Negotiation playbook. Your 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 calls. - Risk tolerance. What constitutes RED for your legal department vs YELLOW. A consumer-facing SaaS company has different RED criteria than a manufacturing firm with concentrated supplier relationships. Configuration is per-org, not template. - NDA triage tiers. Per /triage-nda, 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. - Counsel handoff format. When the plugin escalates to counsel, what context gets passed — full contract plus annotated playbook deviation, or summary plus deviation list. Affects time-to-counsel-decision materially.

First-month configuration time: 8-16 hours of legal-ops time plus 4-8 hours of senior counsel review of the playbook. The Cowork /review-contract overview walks through the configuration pattern.

Step 4: Establish the four use cases first, expand from there

In-house departments that try to deploy Claude across every workflow simultaneously usually produce inconsistent adoption and inconclusive ROI. The cleaner pattern: pick four named use cases, deploy them deeply, expand only after each is in production for 30+ days.

The four starter use cases for most in-house departments:

- NDA triage via /triage-nda. High volume, low complexity, immediate measurable time savings. Most departments process 10-50 NDAs/week; AI triage typically captures 60-80% of standard-tier NDAs without counsel touch. - First-pass contract review via /review-contract on commercial agreements where your team is buyer-side. Saves 30-60 minutes per agreement on the first read; counsel still reviews the flagged sections but starts with annotated context. - Internal memo drafting. Quarterly summaries to leadership, board reporting on legal risk, internal compliance memos. Claude drafts; counsel revises. Saves 40-60% of drafting time on routine memos. - Regulatory landscape monitoring. Specific to industry — for fintech, weekly summary of CFPB and state-level updates; for life sciences, weekly summary of FDA and state pharma updates. Claude drafts the summary; counsel reviews accuracy. Replaces or supplements paid regulatory news services.

The second-order pattern: power users in your department will drift toward use cases 5-15 (clause negotiation against opposing counsel, complex transaction structuring, litigation prep). That's expected and healthy. The discipline is keeping the firm-wide deployment focused on the four-starter set so the ROI memo at year one is defensible.

Step 5: Build the privilege and confidentiality protocol explicitly

In-house Claude deployments inherit different privilege questions than outside firm deployments. The key risks:

- Privilege wave-through. Claude Team and Enterprise carry stronger data-handling commitments than consumer Pro, but the privilege analysis still depends on how the in-house department treats the AI-generated output. Per US v. Heppner (SDNY, Feb 17, 2026), consumer Claude communications were not protected by privilege or work-product doctrine. The Heppner ruling addressed consumer AI specifically; enterprise deployments carry different facts but the privilege analysis is jurisdiction-specific. - Discovery exposure of scratchpad files. The Opus 4.7 multi-session memory feature lets Claude hold context across sessions via scratchpad files. Those files contain matter-specific reasoning. They're discoverable. Storage, retention, and access controls need to be in the protocol from day one. - Attorney-client privilege with the company's outside counsel. When in-house teams use Claude to prepare materials that go to outside counsel, the privilege analysis depends on whether the AI-generated content was prepared at counsel's direction or independently. The Heppner work-product analysis applies here too. Materials prepared without counsel direction may not carry work-product protection. - Cross-border data residency. Multinational legal departments running Claude need to verify which deployment surface satisfies the strictest applicable regime — usually GDPR for EU operations, but increasingly state-level (California, Colorado, Texas) for US deployments. Per Anthropic's pricing, Enterprise tier offers custom data residency terms; Team tier inherits Anthropic's standard posture.

The protocol covers each of these in writing, with named owners for protocol enforcement. The Heppner ruling meets enterprise Claude privilege defense spoke goes deeper on the privilege architecture.

Step 6: Instrument the dashboard before activation

Every in-house Claude deployment that doesn't capture baseline metrics on day one ends up with a defensibility problem at year-one renewal. The dashboard isn't optional infrastructure — it's the artifact that decides whether the legal department renews the procurement contract.

Metrics to capture day one:

- NDA processing time pre-AI vs post-AI. Median minutes per NDA. Track the trend over 30, 60, 90, 180 days. - Contract review cycle time. Time from receipt to first-pass annotated review. The number that goes in the renewal-defense memo. - Counsel touch rate on triaged NDAs. What percentage required counsel review post-triage? If the rate is >30% after 60 days, the playbook configuration needs tightening. - Internal stakeholder satisfaction. Quarterly NPS or equivalent from the business teams the legal department serves. Captures the soft-side ROI that hours-saved metrics miss. - Cost per matter or cost per workflow. Token spend allocated by matter via API logs. Required if the legal department charges back to business units.

The second-order read: the legal department that runs Claude without the dashboard is in the same position as a finance department that doesn't reconcile its accounts. The numbers exist; they just aren't being captured. The dashboard takes about a week of legal-ops engineering time to set up. The cost of not setting it up is the entire defensibility of the original procurement decision.

The Bottom Line: My take: In-house legal departments have the cleanest path to Claude deployment in 2026. Pick Team Standard tier, configure the open-source Cowork plugin against your actual playbook, deploy four named use cases first, lock down the privilege and confidentiality protocol explicitly, instrument the dashboard before activation. A 50-person in-house team lands at $12,000-15,000/year for the model layer plus the plugin — replacing or supplementing legal-ops tooling that historically priced at $50,000+ annual. The structural advantage in-house teams have over outside firms: the budget conversation is internal, and the procurement battle with finance shrinks because the cost folds into a subscription the company likely already has.