Skip to main content
Curia resource · Legal AI risk controls

What can lawyers safely upload to AI tools?

A practical guide for Canadian law firms deciding what information can go into AI, what should stay out of consumer tools, how to supervise staff use, and when a matter-aware workspace is safer.

Confidentiality-first Staff supervision Matter-aware workflow
Quick answer

Do not paste client information into consumer AI by habit.

Lawyers can use AI more safely when the prompt avoids confidential, privileged, personal, and matter-identifying information — or when the firm has approved a controlled workflow that protects that information. The safer default is to treat unredacted matter data as off-limits for casual AI tools.

Product fit

Matter context should stay inside the matter.

A matter-aware workspace helps lawyers avoid ad hoc copy-paste workflows. It gives the firm one place to keep documents, research, drafting, review steps, and AI-assisted output tied to the file instead of spread across general chat sessions.

Explore matter workspaces
Decision rule

Classify the information before you classify the tool.

The first AI-risk question for a law firm is not whether the tool is impressive. It is whether the information being uploaded belongs in that system at all. Professional guidance from the Law Society of Ontario and the Canadian Bar Association points lawyers back to familiar duties: competence, confidentiality, supervision, client communication, candour, and independent judgment.

Consumer AI tools can be useful for low-risk brainstorming, plain-language explanations, and generic drafting support. The risk rises when the prompt includes client facts, privileged analysis, personal information, confidential records, litigation strategy, or file-specific documents.

A practical policy should separate public or generic information from controlled matter data. If the prompt would make a client, matter, witness, settlement position, legal strategy, or confidential document identifiable, it should not be treated as an ordinary chatbot prompt.

Upload risk map

A simple way to decide what belongs in AI.

The exact answer depends on the tool, contract, client, and file. But these categories help a firm set a safer default policy.

  • 01
    Usually lower risk
    Public legal information, generic drafting prompts, non-client hypotheticals, marketing outlines, and administrative text that does not identify a client, matter, witness, strategy, settlement position, or confidential fact.
  • 02
    Needs controls first
    Summaries of pleadings, facts, contracts, correspondence, timelines, medical records, expert materials, or client instructions. Use only where the firm has reviewed the tool terms, security, retention, training-use settings, access controls, and client-communication requirements.
  • 03
    Do not paste into consumer AI by default
    Privileged communications, personal information, settlement strategy, litigation theories, unredacted file records, client names, adverse-party details, confidential business information, and anything a client would not expect to be sent to a third-party AI service.
Firm policy

Four controls before AI becomes routine.

1. Classify the information before prompting. Ask whether the prompt contains client identity, privileged content, personal information, case-specific facts, strategy, or proprietary firm material. If yes, treat it as controlled matter data, not casual AI input.

2. Check the tool before the task. Review the provider terms, data-retention settings, training-use policy, security model, access controls, audit logs, and whether the tool is approved by the firm for client work.

3. Match staff permissions to matter risk. Support staff and students should know which AI tools are approved, what cannot be uploaded, when supervising lawyer approval is required, and how to document AI-assisted work.

4. Keep the lawyer in the review loop. AI output should be checked for confidentiality, accuracy, sources, legal relevance, bias, and whether client or court disclosure is required before it is used externally.

Client communication

Questions to answer before a client asks.

Will the firm use AI only for internal support, or will AI-generated content appear in client-facing or court-facing work?

Will any client information be sent to third-party systems, and if so under what confidentiality, security, and retention controls?

Who reviews AI output before it is relied on?

Does the retainer, engagement letter, or client communication need to address AI-assisted workflows for this file?

This is not a script for every retainer. It is a checklist for deciding when the use of AI is material enough to discuss with the client and document in the file.

Why matter-aware AI is different

The safer workflow is organized around the file, not the prompt box.

Ad hoc AI use pushes lawyers toward a risky habit: extract a few facts from a file, paste them into a general tool, copy the output back, and hope the review trail is good enough. That workflow is hard to supervise and easy to forget.

A matter-aware legal AI workspace changes the operating model. The matter becomes the container for documents, issue lists, research, drafts, and review. The lawyer can inspect how AI-assisted work relates to the file instead of reconstructing a chain of prompts after the fact.

Curia control

Matter-bounded work

Keep research, drafting, documents, and review context organized around the file instead of scattering prompts across consumer chat histories.

Curia control

Reviewable source trail

Tie answers and drafts back to sources, documents, and lawyer review steps so AI output is easier to inspect before reliance.

Curia control

Supervision by design

Give lawyers a clearer place to supervise staff-assisted AI work, compare outputs against the matter record, and preserve judgment at the end of the workflow.

FAQ

Safe AI uploads for law firms.

Can lawyers upload client documents to ChatGPT?
Not safely by default. A lawyer should first assess confidentiality, privilege, privacy, vendor terms, retention, training-use settings, security controls, client expectations, and firm policy. Unredacted matter documents should not be pasted into consumer AI tools casually.
What information is safer to use with AI?
Lower-risk use usually involves public information, generic hypotheticals, non-confidential drafting prompts, and administrative text that does not identify a client or reveal matter-specific facts. Even then, output still needs lawyer review before reliance.
Do lawyers need to tell clients they use AI?
Client communication depends on the role AI plays, the nature of the information involved, the retainer, and applicable professional obligations. If AI affects client work, sends client information to a third party, or materially changes how services are delivered, lawyers should consider whether client communication or consent is required.
How should law firms supervise staff AI use?
Firms should set written AI-use rules, identify approved tools, prohibit unapproved upload of confidential matter data, train staff on redaction and escalation, and require lawyer review of AI-assisted work before it leaves the firm.

Keep AI work tied to the matter.

Curia gives Canadian lawyers a matter-aware workspace for AI-assisted research, drafting, document review, and supervision — with the file context closer to the work and the lawyer still in control.

curia.ca · Toronto, ON
Sources

This resource is general information for legal professionals evaluating AI-assisted workflows. It is not legal advice and does not replace a lawyer’s professional judgment, client-specific confidentiality analysis, or review of applicable rules, court notices, and technology terms.