General-purpose AI is entering legal. What in-house teams should know.

General-purpose AI can now summarize contracts, identify common clause types, and answer plain-language legal questions. For many teams, that's a genuine step forward. Work that used to take 3 to 5 hours of document reading and legal research can happen in under 30 minutes.

But for in-house legal teams handling sensitive documents at volume, the headline capabilities are only part of the picture. Before adopting a generic AI tool for contract review or legal research, several areas are worth evaluating carefully.

What generic AI gets right

General-purpose models have become genuinely useful for a growing set of legal tasks:

• Summarizing long contracts into readable overviews
• Spotting common clause types across different templates
• Answering plain-language questions about legal concepts
• Reducing initial document review time from hours to minutes

This puts AI contract review on every general counsel's agenda. Budgets are opening, pilot programs are launching, and teams that couldn't get approval a year ago are now running evaluations. That's good for the whole legal AI ecosystem.

The question isn't whether generic AI is useful. It's whether it's sufficient for the work in-house teams actually do.

Where the gaps show up

Accuracy and verification. General-purpose models still hallucinate on specialized legal tasks, and confident but unsourced answers often create more work rather than less. Someone still has to verify the output against the actual document. Without citations tied to specific clauses and document versions, AI output is a rough draft that still needs manual review.

In legal work, a 1% hallucination rate can mean a 100% failure. The question is whether the system can show exactly where each answer came from.

Attorney-client privilege. A recent SDNY ruling in US v. Heppner examined whether AI-generated legal documents retain privilege when the provider's privacy policy permits data collection and model training. The finding raised important questions. Most foundation model providers have terms that permit collecting inputs and outputs. In-house teams uploading contracts, employment agreements, or M&A documents to cloud-based AI tools should review how those terms interact with their privilege obligations.

Workflow integration. Generic AI tools operate as standalone chat interfaces. In-house legal teams typically need tools that connect to existing document management systems, route intake requests from business stakeholders, and produce structured output: clause data in tables, playbook deviations flagged, results exportable for compliance reporting. Copying contract text into a chat window and reformatting every response isn't a workflow that scales.

How to evaluate legal AI for your team

If your team is considering AI for contract review, legal research, or intake automation, here are the areas that matter most in an evaluation:

01Data residency and privilege safety

Where does your data go when you use the tool? Can it run on your own infrastructure? Review the provider's privacy policy for language about data collection, model training, and disclosure to third parties. A privilege-safe architecture keeps your documents inside your own environment.

02Citation-grounded answers

Does the system tie every answer to a specific document version and clause? When it can't find a supporting source, does it tell you or guess? At scale, the difference between cited and uncited answers determines whether your team can trust the output without re-reading everything.

03Integration with existing tools

Does it connect to your DMS, email, and collaboration platforms natively? Legal teams adopt tools that fit into how they already work. A tool that requires a separate tab and manual document uploads will struggle with adoption.

04Structured, exportable output

Can it produce clause data in tables, flag deviations from your playbook, and export results for reporting? Outputs that need reformatting before they're useful add time back into the process the tool was supposed to shorten.

05Audit trail and transparency

Is every extraction tied to a source document, version, and timestamp? Can your compliance team inspect how each conclusion was reached? In regulated environments, this is often the deciding factor.

What comes next

General-purpose AI entering legal is a net positive for the industry. It raises the bar for what legal AI tools should deliver and gives in-house teams a reason to start evaluating options they might not have considered a year ago.

The teams that get the most from legal AI will be the ones that look past the surface features and evaluate how the tool handles their data, verifies its answers, and fits into their operations.

Adeu is built around these questions. If you're running an evaluation, we'd be happy to walk through how we approach each one.

Related articles