IRB AI Rules: A Detailed Guide for Counsel and Claimants

The new IRB AI rules will significantly affect how counsel, refugee claimants, appellants, witnesses, and other parties prepare documents for proceedings before the Immigration and Refugee Board of Canada. The rules restrict the use of artificial intelligence in Basis of Claim narratives, affidavits, witness statements, photographs, screenshots, translations, legal submissions, and other materials filed with the IRB.

The Immigration and Refugee Board of Canada has published the formal Practice Notice: Use of Artificial Intelligence in IRB Proceedings. It was signed on July 10, 2026, and comes into force on September 7, 2026.

Although people may casually refer to it as an AI policy, the formal document is a practice notice. It applies to all four divisions of the IRB:

• Refugee Protection Division, or RPD
• Refugee Appeal Division, or RAD
• Immigration Division, or ID
• Immigration Appeal Division, or IAD

The notice applies to represented parties, their counsel, and self-represented parties.

What the IRB AI rules do

The purpose of the practice notice is to protect the integrity of IRB proceedings. The IRB recognizes that artificial intelligence may help people prepare and present their cases. However, AI systems may also produce inaccurate, fabricated, misleading, or unverifiable information.

The central principle is straightforward:

Every party remains responsible for everything submitted to the IRB, regardless of whether artificial intelligence helped prepare it.

Using an AI system does not transfer responsibility to the software provider. Counsel cannot defend an incorrect submission by explaining that ChatGPT, Claude, Gemini, Copilot, or another tool produced the error.

Parties using AI must ensure that the resulting content is:

• Accurate
• Verifiable
• Trustworthy
• Relevant to the particular proceeding
• Supported by reliable sources where appropriate

The IRB also expects parties to be prepared to answer questions about their use of AI. A party may be asked to establish the authenticity and accuracy of a document or explain how it was prepared.

IRB AI rules for BOC narratives

The most important restriction concerns personal evidence.

The practice notice states that parties must not use AI to generate or materially change content that reports or reflects a person’s evidence. This includes:

• Basis of Claim narratives
• Affidavits
• Statutory declarations
• Witness statements
• Personal accounts
• Other documents based on a person’s knowledge or experience

The IRB specifically prohibits using AI to draft a Basis of Claim narrative or rewrite a person’s story.

This means a claimant should not enter personal facts into an AI system and ask it to:

• Write a complete BOC narrative
• Rewrite the claimant’s account
• Make the story more persuasive
• Add emotional language
• Fill gaps in the chronology
• Suggest facts that might strengthen the claim
• Create explanations for inconsistencies
• Generate descriptions of incidents
• Turn interview notes into a first-person narrative
• Make one family member’s narrative sound consistent with another family member’s narrative

Even when the underlying information originally came from the claimant, the resulting document may violate the practice notice if AI generated the narrative or materially changed the claimant’s account.

The claimant must remain the source of the evidence

A BOC narrative is not simply a legal document. It is the claimant’s personal evidence about past experiences, feared harm, state protection, internal flight alternatives, delay, travel history, and other matters relevant to the refugee claim.

The narrative must reflect the claimant’s actual knowledge, memory, language, and experience. It must not become an AI-generated reconstruction of what the system believes a refugee narrative should look like.

This is especially important because an artificial intelligence system may:

• Add details that the claimant never provided
• Replace uncertain memories with precise dates
• Infer motives or intentions
• Exaggerate emotional reactions
• Create artificial transitions between events
• Eliminate inconsistencies without investigating them
• Introduce legal terminology that the claimant does not understand
• Produce language that does not resemble the claimant’s ordinary way of communicating

A polished narrative is not necessarily a reliable narrative. In refugee proceedings, excessive polishing may hide uncertainty, introduce false precision, or distance the claimant from their own evidence.

Can AI correct grammar in a BOC?

The notice permits minor assistive functions such as:

• Spellcheck
• Grammar correction
• Formatting assistance

However, these functions are permitted only when they do not generate the evidence or materially change it.

The distinction is not always easy.

Correcting a spelling mistake is normally minor. Correcting punctuation without changing the meaning is also likely minor. Rewriting an entire paragraph to make it more persuasive, chronological, detailed, or emotionally compelling may be a material change.

Counsel should be particularly cautious with instructions such as:

• “Improve this narrative”
• “Rewrite this professionally”
• “Make this story more persuasive”
• “Add more detail”
• “Make the events flow better”
• “Strengthen the explanation”
• “Fix the inconsistencies”
• “Make this sound credible”

These prompts invite the AI system to do more than correct grammar. They may cause it to create language, reasoning, or facts that did not come directly from the claimant.

A safer approach is to limit any automated review to clearly mechanical corrections. Counsel should then compare the revised document with the original, confirm that no meaning has changed, and review the final version directly with the claimant.

Human preparation of narratives remains possible.

The notice does not prevent counsel or trained staff from helping a claimant prepare a BOC form or narrative.

Counsel may still:

• Interview the claimant
• Ask follow-up questions
• Identify missing information
• Clarify dates and locations
• Organize the account
• Explain the legal issues
• Identify apparent inconsistencies
• Prepare a written narrative based on direct instructions
• Review the completed narrative with the claimant
• Make corrections confirmed by the claimant

The crucial distinction is that a human professional remains responsible for interviewing, understanding, drafting, and verifying the evidence.

Counsel should never assume that the prohibition on AI means claimants must prepare complex narratives without assistance. Careful human assistance remains both permissible and important.

AI-generated affidavits and witness statements

The same restriction applies to affidavits and witness statements.

A witness may provide information orally, through handwritten notes, or in another language. However, counsel should not simply upload those notes to an AI system and ask it to produce a completed affidavit.

The person signing the affidavit or statement must understand and adopt its contents. The document must accurately reflect what the person knows, remembers, observed, or experienced.

Before a witness signs a document, counsel should confirm:

• The witness personally provided the information
• The witness understands every material statement
• The document does not contain AI-generated assumptions
• Dates, names, quotations, and locations have been verified
• Uncertainty has not been replaced with false precision
• The witness is comfortable correcting anything inaccurate
• The final statement is given voluntarily

This is particularly important when the witness does not speak English or French fluently.

AI and photographs, videos, and screenshots

The practice notice prohibits using AI to create or alter:

• Photographs
• Videos
• Screenshots
• Other corroborative evidence

This restriction is broader than a prohibition on completely fabricated images. It can also affect AI-based image enhancement and editing.

Potentially problematic activities include:

• Generating a new photograph
• Adding or removing a person
• Replacing a background
• Removing an object
• Reconstructing a damaged area
• Using generative fill
• Modifying text inside a screenshot
• Enhancing a face
• Changing lighting in a way that affects interpretation
• Removing shadows, reflections, watermarks, or markings
• Creating missing visual details through AI upscaling
• Combining separate photographs
• Colourizing an image where colour could affect its meaning
• Altering metadata
• Producing a simulated version of an unavailable document

An AI enhancement tool may create pixels that were not present in the original file. The resulting image may appear clearer, but it may no longer be a faithful copy of the original evidence.

What about cropping or rotating an image?

The practice notice does not provide detailed rules for every technical adjustment.

Ordinary operations such as rotating a sideways image, converting a file to PDF, reducing its file size, or cropping empty borders may be administrative rather than evidentiary changes. However, parties should proceed cautiously.

A prudent practice is to:

1. Preserve the original file.
2. Avoid modifying the evidentiary content.
3. Keep an unedited copy available.
4. Explain any necessary technical processing.
5. Submit the original when authenticity may be disputed.
6. Avoid AI-based enhancement tools altogether.

Counsel should also obtain screenshots and digital files directly from the claimant whenever possible. Repeated forwarding, editing, or conversion may remove metadata or reduce the ability to establish authenticity.

When an AI declaration is required

When AI generates or materially changes content in a document submitted to the IRB, the party must include the declaration prescribed by the practice notice.

Artificial intelligence (AI) was used to create or materially change text in this document. All content generated by AI, and the authenticity of any cited case law or other legal authorities, has been reviewed and verified by [name].

The declaration must disclose that AI was used and identify the person who reviewed and verified:

• The AI-generated content
• The authenticity of cited decisions
• The authenticity of other legal authorities

Parties may also identify:

• The software used
• The purpose for which it was used
• Whether it affected the whole document
• Whether it affected only particular paragraphs or sections

The exact required declaration appears in the official IRB practice notice. Parties should use the IRB’s wording rather than inventing a shorter or less precise version.

A declaration is not required where AI was used only for minor spellcheck, grammar correction, or formatting and did not generate or materially change the wording.

When uncertain, disclosure is generally safer than silence. However, disclosure does not make a prohibited use permissible. A party cannot use AI to draft a BOC narrative and then cure the problem merely by declaring the AI use.

AI translation and transcription

AI translation and transcription require particular care.

The general practice notice states that the use of AI for translation or transcription must always be disclosed. It also reminds parties that the rules of each division require translated documents to be accompanied by a declaration from a human translator.

However, disclosure alone does not necessarily make an AI translation acceptable.

For documents submitted to the RPD and RAD, the IRB’s electronic document filing instructions state that a human must perform the translation. The instructions say not to use AI tools or web services.

The RPD also states that it will not accept documents translated through software, AI, or services such as Google Translate.

Therefore, counsel and refugee claimants should not rely on an AI-generated translation for RPD or RAD evidence.

A compliant translation should include:

• A copy of the original document
• A complete English or French translation
• The translator’s name
• The original language and dialect, if applicable
• A declaration confirming the translation’s accuracy
• A signature and date where required

AI may be useful for helping counsel understand the general subject of a document before obtaining a formal translation. That internal convenience should not be confused with an acceptable evidentiary translation.

AI in legal research and submissions

The IRB AI rules do not impose a complete ban on using artificial intelligence for legal work.

AI may assist with matters such as:

• Generating an initial research plan
• Identifying potential legal issues
• Organizing written submissions
• Summarizing counsel’s own notes
• Editing non-evidentiary correspondence
• Preparing an outline
• Comparing arguments
• Suggesting headings
• Reviewing a document for repetition
• Helping counsel identify questions requiring further research

However, counsel remains responsible for the final submission.

Every legal proposition, statutory provision, regulatory reference, policy statement, and case citation must be verified through a reliable source.

Reliable sources may include:

• Official government websites
• CanLII
• Court websites
• Recognized legal databases
• Commonly used commercial legal publishers

An AI-generated case summary is not itself a reliable legal source.

Hallucinated decisions remain a serious risk.

Artificial intelligence systems may produce:

• Non-existent cases
• Incorrect neutral citations
• Incorrect paragraph references
• Misidentified courts
• Invented quotations
• Incorrect summaries of legal tests
• Decisions that do not support the stated proposition
• Outdated versions of legislation or policy
• Cases from the wrong jurisdiction

Counsel should open and read every cited authority. Searching for a case title alone is not enough. Counsel must confirm that:

1. The decision exists.
2. The neutral citation is correct.
3. The quoted passage appears in the decision.
4. The cited paragraph supports the stated proposition.
5. The decision has not been overturned or materially distinguished.
6. The legal principle applies to the facts and division involved.

Where AI generated or materially changed the submission, the required AI declaration must also be included.

Do not use AI to inflate submissions.

The practice notice warns parties not to fill submissions with AI-generated information unrelated to the particular facts of the case.

Longer submissions are not necessarily stronger submissions.

AI can generate many pages of apparently sophisticated analysis within seconds. However, generic passages about the refugee definition, credibility, state protection, internal flight alternatives, humanitarian factors, detention principles, or procedural fairness may distract from the actual issues.

Effective submissions should:

• Identify the determinative issues
• Connect the law to the evidence
• Address weaknesses honestly
• Use only relevant authorities
• Avoid repeating the same proposition
• Distinguish general country conditions from the claimant’s circumstances
• Assist the member rather than overwhelm the record

Counsel should treat AI output as a draft requiring judgment, not as completed advocacy.

Privacy and confidentiality risks

The practice notice also tells parties to consider privacy risks.

Refugee and immigration files may contain highly sensitive information, including:

• Identity documents
• Passport information
• Addresses and telephone numbers
• Medical and psychological records
• Sexual orientation or gender identity information
• Religious or political beliefs
• Allegations of persecution
• Information about family members abroad
• Police and military records
• Information about minors
• Details about smugglers, agents, or criminal organizations
• Confidential communications
• Personal photographs and videos

Uploading this information to a public or consumer AI service may create confidentiality, privacy, data retention, or security concerns.

Before using any AI tool, counsel should understand:

• Whether prompts are stored
• Whether data may be used for model training
• Where the data is processed
• Whether the service provides enterprise privacy protections
• Whether other users can access the information
• Whether the information can be deleted
• Whether the tool complies with counsel’s professional obligations
• Whether the client has provided meaningful and informed consent

Removing a person’s name may not be enough. A detailed combination of dates, locations, family relationships, political activities, and incidents may still identify the claimant.

Consequences of violating the practice notice

Failure to follow the IRB AI rules may have serious procedural and professional consequences.

The IRB may:

• Refuse to accept a document for filing
• Decline to rely on the document
• Draw negative credibility inferences
• Report counsel’s conduct to the relevant professional regulator
• Restrict counsel from appearing before the IRB
• Prohibit counsel from appearing before the IRB
• Take any other action considered necessary

The credibility consequence is particularly important in refugee proceedings.

If a member concludes that a narrative, affidavit, photograph, or screenshot was generated or materially altered by AI, the concern may extend beyond that individual document. It may affect the member’s assessment of the reliability of the party’s other evidence.

A failure to disclose AI use may also create a separate concern about transparency and candour.

Advice for counsel

Counsel appearing before the IRB should establish a clear internal AI protocol before the practice notice takes effect.

1. Adopt a written office policy

The policy should identify:

• Permitted AI tools
• Prohibited tools
• Permitted purposes
• Prohibited purposes
• Information that must never be uploaded
• Verification requirements
• Disclosure requirements
• Who may authorize AI use
• How AI use will be recorded
• How original evidence will be preserved

Staff, students, contractors, interpreters, translators, and co-counsel should receive the same instructions.

2. Use a human-first narrative process

The preparation of personal evidence should begin with the claimant, not with an AI prompt.

A sound process may include:

1. Conducting a detailed human interview.
2. Asking open-ended and follow-up questions.
3. Identifying uncertainty and inconsistency.
4. Obtaining clarification directly from the claimant.
5. Preparing the narrative through human drafting.
6. Reviewing the draft with the claimant in a language they understand.
7. Recording corrections requested by the claimant.
8. Obtaining final confirmation before filing.

AI should not replace the interview or become the author of the claimant’s personal history.

3. Give claimants written instructions

Claimants may independently use AI without telling counsel.

At the beginning of the retainer, counsel should advise clients not to use AI to:

• Draft or rewrite their BOC
• Prepare a witness statement
• Create an affidavit
• Translate evidence
• Edit photographs
• Enhance screenshots
• Invent missing details
• Research answers they think the IRB wants to hear

Counsel should ask clients to disclose any AI use that occurred before the retainer.

4. Keep an AI-use record

For each file, counsel may wish to record:

• Name of the tool
• Date of use
• User who operated the tool
• Purpose of use
• Documents affected
• Information entered
• Verification completed
• Whether a declaration is required
• Name of the person who verified the output

This record may help counsel respond if the IRB later asks how a document was prepared.

5. Preserve original evidence

Keep original versions of:

• Photographs
• Screenshots
• Audio recordings
• Videos
• Emails
• Messages
• Witness notes
• Handwritten statements
• Draft narratives
• Source documents
• Human translations

Do not overwrite original files with edited versions.

6. Verify every authority

Assign a named person to verify each case, statute, regulation, policy, and quotation. That person should verify the source directly, not through another AI summary.

7. Review current files

Counsel should not wait until September 7, 2026, to reconsider their workflows.

Review pending matters that may remain before the IRB after the effective date. Identify whether AI was used to prepare:

• BOC narratives
• Affidavits
• Witness statements
• Photographs
• Screenshots
• Translations
• Memoranda
• Written submissions

Where a possible problem exists, counsel should assess the appropriate corrective action before filing further material or appearing at the hearing.

The appropriate response will depend on the facts. It may involve correcting evidence, providing an amended document, making disclosure, obtaining a human translation, or seeking legal advice about professional obligations.

Advice for refugee claimants

Refugee claimants should understand that AI may produce fluent language while still changing the truth of their account.

Claimants should:

• Tell their story in their own words
• Identify when they do not remember a date
• Avoid guessing
• Review every page of the BOC
• Correct errors before signing
• Ask for an interpreter when needed
• Keep copies of drafts and final documents
• Preserve original photographs and messages
• Tell counsel about any previous AI use
• Avoid uploading confidential evidence into public AI systems
• Never allow AI to invent missing facts

A claimant should not sign a BOC form, affidavit, or witness statement merely because the document sounds professional.

Before signing, the claimant should be able to explain and confirm every material statement in the document.

Advice for witnesses and family members

Witnesses and family members should prepare their evidence independently and truthfully.

They should not use AI to make their statements match the claimant’s narrative. Genuine witnesses may remember events differently. Minor differences do not necessarily mean that someone is lying.

Artificially harmonizing statements can create greater credibility problems than the original differences.

Each witness should:

• Describe only what they personally know
• Distinguish direct knowledge from information received from others
• Identify uncertainty
• Avoid copying another person’s statement
• Review the final document carefully
• Correct anything they do not understand
• Disclose any use of AI

Practical checklist under the IRB AI rules

Before submitting a document to the IRB, counsel and parties should ask:

• Does this document report a person’s evidence?
• Did AI draft any part of the person’s account?
• Did AI rewrite or materially change the wording?
• Did AI add any facts, dates, explanations, or emotions?
• Was a photograph, video, or screenshot altered?
• Was AI used for translation or transcription?
• Do the applicable division rules permit the translation?
• Does the document require an AI declaration?
• Has every case citation been opened and verified?
• Has every quotation been checked?
• Is the content relevant to the particular case?
• Was confidential information entered into an insecure system?
• Can the party explain how the document was prepared?
• Has the original evidence been preserved?
• Has a named human reviewer accepted responsibility for the final document?

If the answer reveals prohibited or undisclosed AI use, the issue should be addressed before filing.

The importance of professional judgment

Artificial intelligence can support legal work, but it cannot assume professional responsibility.

The new practice notice does not require counsel to reject technology. It requires counsel to use technology in a way that preserves:

• Authenticity
• Accuracy
• Confidentiality
• Transparency
• Evidentiary integrity
• Professional accountability

The most important distinction is between assistance and substitution.

AI may assist counsel with legal research, organization, and limited editing. It must not substitute for the claimant as the source of personal evidence, for the witness as the source of testimony, for the translator as the person responsible for a translation, or for counsel as the professional responsible for advocacy.

Conclusion

The IRB AI rules represent a significant change in the preparation of immigration and refugee proceedings.

The practice notice prohibits using AI to draft or materially rewrite BOC narratives, affidavits, witness statements, and other personal evidence. It also prohibits using AI to create or alter photographs, videos, screenshots, and corroborative evidence.

AI may still assist with legal submissions and other non-personal materi. Still, parties must verify the content, use reliable sources, protect confidential information, avoid irrelevant bulk, and provide the required declaration when AI generates or materially changes a document.

Counsel should update their office procedures, train staff, advise clients, preserve original evidence, document AI use, and review pending files before September 7, 2026.

Claimants should remember one essential principle: their BOC and personal evidence must remain their story, based on their own knowledge and experience.

Official source: Practice Notice: Use of Artificial Intelligence in IRB Proceedings

This article provides general information and does not constitute legal advice. The application of the practice notice may depend on the facts, the type of document, the nature of the AI use, and the division hearing the matter.