The Brief:
Freshfields has partnered with Anthropic to co-build legal AI, becoming the first Magic Circle firm to publicly back a bespoke, multi-vendor strategy over a single specialist platform.
The deal sharpens a growing divide in BigLaw between firms building their own AI and those going all-in on Harvey or Legora.
Everyone thought law firms would buy AI off the shelf. Most have. Freshfields just chose a different path.
The Magic Circle firm has struck a multi-year deal with Anthropic to jointly build legal AI tools, rolling Claude out across all 33 offices to 5,700 staff.
It is the first partnership of its kind Anthropic has made with a law firm, and the clearest signal yet that BigLaw’s AI strategy is forking.
The Freshfields play
As part of the Freshfields’ deal:
Claude is being deployed firmwide, and Freshfields will gain early access to future Anthropic models and tools.
A co-development programme will see Freshfields lawyers and Anthropic engineers build products for document drafting, contract review and due diligence.
Freshfields is paying an undisclosed sum to Anthropic, and Anthropic cannot use Freshfields’ data to train its models.
Chief Innovation Officer Gil Perez put the rationale plainly.
It’s a really inefficient and shortsighted approach to say I’m going with a single vendor because things are changing so fast. So tomorrow when Gemini comes up with a better model, we’ll use that, when OpenAI comes with a better model, we use that.
That is a pointed comment. Because most of BigLaw has done the exact opposite.
The buy camp
Harvey, now valued at US$11bn with around US$190m ARR, has locked in global majors like Latham & Watkins and Macfarlanes. Legora, valued at US$5.55bn after its Accel-led Series D in March, has the likes of Linklaters and White & Case.
The pitch is simple.
Both platforms are purpose-built for legal work, deeply embedded in existing workflows, and developing fast.
For most firms, buying a best-in-class legal platform is the lowest-friction path to AI adoption.
The build camp
A handful of firms have gone further.
A&O Shearman built agentic AI tools with Harvey covering antitrust, fund formation and loan review, then moved to sell them back to clients and rival firms under a subscription model, retaining a slice of revenue. Cleary Gottlieb went further still, acquiring gen AI company Springbok in March 2025 to build custom solutions in-house.
The logic is differentiation and flexibility.
Build your own, and you can pick the best model for each task rather than being locked into one provider’s roadmap.
Freshfields’ Gerrit Beckhaus put it this way: “We basically integrate the gen AI models on various steps of the production chain. There it gives us a lot of flexibility to choose the right model for the right task.”
The counterargument is economics.
Maintaining a genuine in-house AI engineering function is expensive and difficult. Bloomberg’s reported experience with proprietary BloombergGPT, spending over US$10m before finding off-the-shelf GPT-4 outperformed it.
Australian approach
Australia’s top tier has already split. Harvey has won Ashurst, Mallesons, G+T, Clayton Utz and Corrs. Legora has won MinterEllison and Allens. HSF Kramer has adopted both.
But Freshfields’ move may add another variable. Foundation-model vendors are now moving directly into the workflow layer that Harvey and Legora have owned.
Whether bespoke builds deliver a lasting edge, or whether firms that backed Harvey or Legora early made the smarter bet, is still an open question. And whether any Aussie firms follow Freshfields down its path is a question worth watching.
Source: Financial Times, Law.com