Courts across Canada have already made their expectations clear on that front: lawyers remain responsible for the materials they file, whether AI was involved in drafting them or not.

A more interesting question is beginning to emerge.

What happens when the conversation with the AI becomes evidence?

Organizations are increasingly using Microsoft Copilot, ChatGPT, Claude, and other generative AI tools to brainstorm ideas, summarize information, evaluate options, and draft communications. Each of those interactions creates a digital trail: prompts, chat histories, uploaded documents, generated outputs, and audit records.

In litigation, those records may prove just as significant as the final document that was ultimately created.

The prompt itself may reveal who knew about a problem, what risks were considered, what alternatives were explored, and how a decision evolved over time.

That possibility creates new questions for litigants and organizations alike.

Are AI prompts discoverable?

Are AI chat histories covered by legal hold obligations?

What happens when privileged information is entered into a third-party AI platform?

Canadian courts have not yet provided definitive answers to many of these questions. However, existing discovery principles suggest that AI-generated records will not remain outside the evidentiary landscape for long.

Key takeaway

The greatest litigation risk associated with AI may not be inaccurate answers.

It may be the permanent record of the questions that were asked.

Organizations that are implementing AI should begin thinking about preservation, retention, privilege, and governance now, before those issues arise in active litigation.

As with many technology-driven legal developments, the underlying legal principles are familiar. What is changing is the volume, format, and visibility of the information those principles will need to govern.