Why Law Firms Need More Than Direct LLM Access
Vasu Aggarwal - Co-founder
There is a tempting view inside law firms that AI strategy can be reduced to a simple procurement question: buy ChatGPT, Claude, Gemini, or whichever model is most talked about this quarter. It is understandable. The underlying models are powerful, fast-moving, and increasingly impressive.
But legal AI adoption is not really a model-selection problem. It is a workflow, verification, governance, and adoption problem.
That distinction matters because the model market moves in cycles. Freshfields is a useful example. In April 2025, the firm announced a strategic collaboration with Google Cloud to roll out Gemini and build legal AI products across the firm. In April 2026, Freshfields announced a major partnership with Anthropic to deploy Claude globally and co-build legal workflows.
That does not mean one provider won and another lost. It means the market is changing too quickly for most firms to build their legal AI strategy around a single model subscription. The better approach is a unified legal interface that sits above the model layer, chooses the right provider for the right task, and can switch as models improve, degrade, become unavailable, or become commercially unattractive.
That flexibility also reduces operational risk. Claude has had recent service incidents, including an April 28, 2026 disruption affecting Claude.ai and API access. For a lawyer working against a filing deadline, the question is not whether Claude, Gemini, or GPT is best in the abstract. The question is whether the work can continue reliably.
In legal work, citations are not a feature. They are the bridge between usefulness and professional liability.
The second issue is verification. Legal work cannot rest on an answer that merely sounds right. Lawyers need to see where an answer came from, trace it back to the document, and verify the underlying source. If an associate uploads a bundle, asks for a chronology, and receives a confident answer without click-through support, the review burden simply shifts from drafting to forensic checking.
This is where legal-specific platforms should be materially different from direct LLM usage. Uploaded documents should produce highlighted citations. Research answers should show the path back to legislation, definitions, consequences, authorities, and judicial treatment. A user should be able to move from answer to source without guessing.
The third issue is cost and control. Direct subscriptions can look cheaper until usage becomes real. Teams start with a simple monthly plan, then encounter message limits, credit purchases, tool restrictions, and inconsistent availability during the working day. The sticker price is not always the total cost of ownership.
Security is similar. The wrong plan can create the wrong risk posture. Claude Free, Pro, Max, Teams, Enterprise, API access, and cloud deployments are not interchangeable from a data governance perspective. Anthropic's own materials distinguish consumer products from commercial products, and consumer retention settings can differ depending on whether users opt into data use for model training. For law firms, that complexity is not a footnote. It affects confidentiality, retention, privilege, cross-border storage, and client commitments.
Finally, there is adoption. Most lawyers will not become expert AI operators just because the firm buys them access to a frontier model. They need training, prompt patterns, matter-specific workflows, guardrails, escalation paths, and examples that map to the work they already do. Without that service layer, usage clusters around a few tech-forward lawyers while the rest of the firm remains cautious, confused, or underwhelmed.
The real differentiation is not "we have AI and they have AI." Everyone has AI now. The differentiation is whether the platform removes model-selection burden, preserves flexibility across providers, verifies outputs against source material, manages security and governance, supports legal research methodology, and helps lawyers actually adopt the technology in daily work.
Direct LLMs are extraordinary engines. But law firms do not need an engine sitting in the middle of the office. They need a working vehicle, with steering, brakes, mirrors, insurance, and someone to help the team learn how to drive it properly.
Sources
- Freshfields and Google Cloud strategic collaboration, April 2025
- Freshfields and Anthropic partnership announcement, April 2026
More Stories
