Professionals who carry personal liability for their decisions — lawyers, tax accountants, auditors — cannot afford AI that gets it wrong. As enterprises accelerate deployment of agentic systems, the firms serving those professionals are discovering that fiduciary grade AI — intelligence built on governed, authoritative data — is not a constraint on adoption; it is the very thing that makes high-stakes AI possible.

Thomson Reuters Corp., which has built its enterprise AI and data platform on Snowflake to deliver trusted intelligence at scale, is demonstrating what fiduciary grade AI looks like in practice. The company’s data estate — spanning more than 37,000 governed tables and 350 databases — provides the foundation for AI tools its legal, tax and audit customers can stake their reputations on, according to Bala Kasiviswanathan (pictured, right), vice president of developer and AI experiences at Snowflake Inc.

“All these tools for AI are not real until they are on a governed data platform,” Kasiviswanathan said. “Thomson Reuters is a great proof of how they spent years building that trusted governed data foundation — over 37,000 tables, 350 databases. And that foundation allows them to do some amazing things, and at a speed at which they want to innovate.”

Kasiviswanathan, Caitlin Halferty (center) head of data and analytics at Thomson Reuters Corp., and Laura Safdie (left) head of legal innovation at Thomson Reuters Corp. spoke with  theCUBE’s Dave Vellante at Snowflake Summit 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed fiduciary grade AI, the evolution of CoCounsel and how a governed data estate accelerates rather than constrains enterprise AI. (* Disclosure below.)

Fiduciary grade AI demands authoritative data and verifiable outputs

For Thomson Reuters, fiduciary grade AI is not a marketing phrase — it is an operational standard built on four components: authoritative content, investment in data security and safety, subject matter experts who vet and validate outputs, and transparent results. The concept is most visible inside CoCounsel, Thomson Reuters’ legal AI assistant, which Safdie co-created before the company acquired Casetext in 2023 for $650 million.

The concept is most visible inside CoCounsel, Thomson Reuters’ legal AI assistant. As one of CoCounsel’s architects, Safdie is direct about what fiduciary grade AI requires. “You can’t work with AI that isn’t rooted in verifiable legal data,” she said. “The difference between using AI that’s unconnected to legal data, unconnected to legal expertise can mean getting it wrong. And so we work with the professionals who need to get it right every time.”

CoCounsel has evolved from an early retrieval-augmented generation approach into a fully agentic assistant natively built on Thomson Reuters’ authoritative legal content, including Westlaw and Practical Law. Safdie explained that rebuilding the assistant around an agentic harness unlocks a complex work product that earlier generations could not deliver — and critically, keeps the lawyer in charge of every step. At the same time, Thomson Reuters’ responsible AI team vets every capability for hallucination and bias before it ships, Halferty noted.

“What’s important about the lawyers that you work with, it’s their human lawyering,” Safdie said. “An AI is not going to bring that human layer. What we’re able to do is see lawyers adopt AI for the more mechanical parts of practicing law. Leaving the lawyer to be the one that connects with you, that stands with you in court”.

The same fiduciary grade standard governs Thomson Reuters’ internal data transformation, Halferty noted. By building a semantic capability on Snowflake — unifying more than 23 previously fragmented customer master data sources — the company has given its finance teams a single, trusted definition of core business terms. Snowflake’s CoWork and CoCo tools extend that governed foundation to more than 1,500 users, Kasiviswanathan added, allowing business users to query data in natural language and builders to develop new capabilities without losing track of what each data point means.

“Self-serve is the future”, Kasiviswanathan said. “Once you have that kind of secure governed platform, you can actually have people go and do what they need to do and remove all sorts of bottlenecks so that you can move fast”.

The partnership also points toward where Thomson Reuters is heading next: enterprise-scale semantic intelligence as the substrate for a new generation of AI agents. Safdie closed with the clearest statement of where the company is heading.

“You can’t just bring AI to an industry you don’t know, you can’t just transform a workflow you don’t understand,” she said. “If you are building for one of the most important professions in the world, you need to understand how AI can make us better and serve the public interest, while also doing it in such a way that’s fit for purpose.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the Snowflake Summit 2026:

(* Disclosure: TheCUBE is a paid media partner for Snowflake Summit 2026. Sponsors of theCUBE’s event coverage do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE