CoCounsel: AI output still needs an accountable review
· BehindLines newsroom
Thomson Reuters announced a new generation of CoCounsel Tax, Audit & Accounting on October 22, 2026. According to the company, it is available to its customers in the United States. The supplier describes software that prepares calculations and documents for a professional to review.
SiliconANGLE reports on a vendor briefing and demonstration. External reporting on that presentation is not an independent reliability test of the product.
Producing work and checking it
The announcement suggests a different question to ask of artificial intelligence. An answer can sound convincing; reviewable work must also allow someone to examine how it was produced. That means distinguishing the consistency of a calculation, the choice of data and the suitability of assumptions.
Consider an abstract example: correctly adding an unsuitable input still produces an unusable result. Making an operation visible helps explain what happened, without deciding whether that operation was appropriate to the problem. This reasoning shows why professional review continues to matter.
What a demonstration cannot measure
Assessing a promise of reviewable work would require cases that were not selected for a presentation: missing information, ambiguous instructions or conflicting documents. It would also require examining how mistakes are detected and corrected, and comparing the effort of checking the output with preparing the work manually.
These are evaluation questions, without findings established here about CoCounsel. Neither the announcement nor the account of the demonstration provides, on its own, an independent measure of general accuracy or time saved.
The software's interest therefore lies in the working relationship it proposes: producing something a professional can question before accepting it. Its announced availability concerns U.S. customers.