Legal AI is changing its stack. Thomson Reuters trained its own AI model, then kept using Anthropic’s model anyway — a useful reminder that owning the model is not the same as owning the workflow.
That tension now sits beside a larger enterprise push. On August 20, 2026, Thomson Reuters and iManage announced an expanded strategic partnership to deliver AI-powered legal workflows with governance features, connecting Thomson Reuters AI, legal content, and workflow solutions with governed organizational knowledge in the iManage platform.
The collaboration brings CoCounsel Legal into the iManage environment and integrates with HighQ, Noetica, and Legal Tracker. The integration spans AI-assisted legal work, transactional workflows, and matter management, turning the partnership into something more practical than another model announcement dressed in enterprise clothing.
Context Matters More Than Model Ownership
The partnership adds Model Context Protocol, or MCP, support. Approved Thomson Reuters AI tools will be able to reason from governed iManage content while preserving access controls, ethical walls, and privilege boundaries — the guardrails that matter when legal knowledge moves through an AI workflow.
Key capabilities include drafting assistance grounded in authoritative legal content, MCP-enabled access to governed matter content, connected transactional workflows, connected matter intelligence, and knowledge that stays connected as work moves. API-based integrations between the iManage platform and various Thomson Reuters solutions are available today, while MCP support is coming soon.
Rawia Ashraf, Co-Head of CoCounsel Legal at Thomson Reuters, framed the problem around fragmented work. She said legal work lives in too many places and described the renewed partnership as a way to connect CoCounsel Legal, HighQ, Noetica, and Contract Express, with documents flowing from iManage into the workflow, work product flowing back automatically, and the connection staying current as matters evolve.
Ashraf also said legal teams have spent years building their knowledge in iManage and that Thomson Reuters wants that knowledge to work across everything it builds. That is the real pitch: not another isolated assistant, but a route for existing organizational knowledge to travel through more legal processes without losing its controls.
Harvey Builds Its Own Exit Ramp
Harvey introduced Harvey Tenet, its first in-house proprietary AI model for legal work, on August 18, 2026. Gabe Pereyra, cofounder of Harvey, now has a model designed inside the company after Harvey built an $11 billion legal-software business on top of other companies’ AI models.
That sequence matters. Harvey Tenet is not arriving in a vacuum; it follows a business built with external models and marks a move toward proprietary technology for legal work. Thomson Reuters, meanwhile, trained its own model but continued using Anthropic’s AI model, showing that enterprise AI strategies can combine internal systems with outside models rather than choosing one camp.
Ryan Begin, Vice President, Technology Partnerships and Ecosystem Strategy at iManage, said legal work depends on bringing the right sources of knowledge together in the right context. He described the partnership as a combination of Thomson Reuters’ authoritative legal content, AI, and workflow capabilities with the governed knowledge managed in iManage.
Begin also pointed to the need for organizational context and controls as AI takes on more complex work. That is where the iManage integration earns its keep: the model does not simply receive more information; approved tools access governed content while the platform preserves access controls, ethical walls, and privilege boundaries.
The contrast between Harvey Tenet and Thomson Reuters’ continued use of Anthropic’s model reveals the market’s less glamorous truth. Legal technology companies can train proprietary models, rely on external models, or use both — but customers still need connected workflows, authoritative content, and governance that survives contact with actual matters.
Model ownership gets the announcement. Controlled access to useful knowledge gets the work done.
Based on
- Thomson Reuters trained its own AI model. Then it kept using Anthropic’s anyway. — thenewstack.io
- iManage and Thomson Reuters Announce Strategic Partnership to Deliver Governed AI-powered Legal Workflows | Markets Insider — markets.businessinsider.com
- Harvey Announces Tenet, Its First in-House AI Model for Legal Work – Business Insider — businessinsider.com
