Canada

AI Regulations

With respect to existing AI regulations, at the federal level, the Government of Canada’s Directive on Automated Decision-Making is directed to the use of AI by the federal government to ‘make or support administrative decisions to improve service delivery’. The directive mandates that any such automated decision system must complete an algorithmic impact assessment with respect to the use of technology. This requires an identification of the impact level of any such system and the appropriate mitigation measures that should be adopted.

The Government of Canada has also issued a Guide on the use of generative AI in June 2025, endorsing alignment with ‘FASTER’ (fair, accountable, secure, transparent, educated and relevant) principles that were developed by the Treasury Board of Canada Secretariat as to how institutions and public servants should use generative AI in their work.

At the provincial level, Ontario’s Responsible Use of Artificial Intelligence Directive applies to all Ontario government ministries and provincial agencies, or to any AI system or service that is otherwise used in the development, delivery or decision-making of a government policy, program or service. Similar to the approach adopted federally, the directive adopts a ‘risk management’ approach which requires appropriate controls based on the risk level, as well as disclosure and reporting.

Guidelines for practitioners

As at June 2026, no public source has confirmed that Canada has formally adopted the UNESCO Guidelines for the Use of AI Systems in Courts and Tribunals.

Canadian Judicial Council

The Canadian Judicial Council, which oversees the conduct of federally appointed judges, published the first edition of its Guidelines for the Use of Artificial Intelligence in Canadian Courts in September 2024. The guidelines are centred on three ‘guiding principles’, namely, (i) awareness and education; (ii) caution; and (iii) leadership. The guidelines do not themselves prohibit the use of AI, noting that some forms of AI have already been embedded in tools currently used for everyday judicial applications such as translation and legal research. However, the Guidelines stress awareness of risks, prevention of delegation of judicial decision-making authority, and safe, effective and appropriate use of AI by the judiciary. The guidelines also point to the need for a more standardised approach to judicial education in AI.

Federal Court of Canada, Interim Principles and Guidelines on the Court’s Use of AI (2025)

The Federal Court of Canada, a national trial court which decides disputes in the federal domain, published Interim Principles and Guidelines on the Court’s Use of Artificial Intelligence in September 2025. The Court states that it will not use AI, and more specifically automated decision-making tools, in making judgments or orders without first engaging in public consultations. The Court also stated that AI tools are being used to assist language specialists who translate Federal Court decisions, as complementary tools that do not replace the human doing the work and with quality control in place.

Royal Canadian Mounted Police, National Technology Onboarding Program Transparency Blueprint (2024)

The Royal Canadian Mounted Police (RCMP)’s National Technology Onboarding Program Transparency Blueprint acknowledges the multiple uses of AI by law enforcement and identifies various considerations with respect to their use. These include, inter alia, crime forecasting, classifying and editing photo and video evidence, gunshop detection and transcription and translation. The RCMP blueprint reiterates principles such as accountability measures to ensure oversight of AI systems, the need to avoid continuing existing biases and discriminatory practices as well as respect for privacy rights in their use of AI systems. It also refers to the existence of “Interim Guidance on Generative Artificial Intelligence Tools”, said to be applicable “while an official AI policy is being finalised”.

Toronto Police Service, AI Technology Policy (2024)

Toronto Police Service’s AI Technology Policy (last amended 11 January 2024) is a rare exception to the general lack of openly disclosed policies of provincial law enforcement authorities. The policy identifies a number of guiding principles which apply to all use of technology including AI technology including (among others) compliance with all applicable law, fairness (i.e. not increasing or perpetuating bias in policing) and transparency. The policy provides that new AI technologies will only be adopted in consultation with a number of stakeholders which includes independent human rights, legal and technology experts and affected communities, and identifies a ‘risk category’ for such technologies and appropriate risk mitigation measures and assessment before tools are used and deployed. Following deployment, AI technology will be monitored and subject to continuous review.

Provincial and Territorial Bar Associations and Law Societies

Nearly every single provincial bar has issued specific guidance to their licensees on the use of AI, which to date has centred primarily on generative AI use. The guidance across the various provinces and territories reiterate the application of existing professional obligations under the Federation of Law Societies of Canada’s Model Code, including of competence and technology competence and confidentiality, as well emphasising the need for independent verification of AI outputs and appropriate supervision.

Individual courts, too, are setting out their expectations of both legal professionals and unrepresented litigants with respect to AI. These have taken the form of notices or Court practice directions. Courts in Alberta, Manitoba (Court of King’s Bench), Ontario (Superior Court of Justice), Nova Scotia (Supreme Court of Nova Scotia and Nova Scotia Provincial Court), Newfoundland and Labrador (Supreme Court of Newfoundland and Labrador), the Northwest Territories (Supreme Court of the Northwest Territories) and the Yukon (Supreme Court of the Yukon) have all issued such notices or directions. The Federal Court of Canada has also issued a notice to similar effect.

Notwithstanding the proliferation of rules and guidance, expectations of litigants are not always consistent across jurisdictions. . Some courts, like in the Yukon and Manitoba and the Federal Court, explicitly require lawyers to disclose the use of AI in connection with the preparation of submissions. Others, like Alberta, do not.

As for Crown prosecutors, they are subject to the relevant regulatory guidance by their respective regulators. But in addition, as employees of the Government, federal Crown counsel are also subject to the Government of Canada’s Guide on the Use of Generative AI (see above). Similar directives apply to Crown attorneys or counsel in certain provincial jurisdictions where government employees have been issued guidance on the use of AI technologies (such as Ontario’s Responsible Use of Artificial Intelligence Directive).

Criminal procedure rules

The ordinary rules of criminal procedure in Canada apply to evidence that is AI-generated or otherwise analysed using AI. This includes common law principles of evidence, constitutional protections under the Canadian Charter of Rights and Freedoms, and criminal legislation governing the admission of evidence.

Under Canadian common law, the governing rule of evidence is that evidence relevant to a fact in issue is admissible unless an exclusionary rule applies, and subject to the Court’s residual discretion to exclude evidence where its prejudicial effect substantially outweighs its probative value. For instance, AI-generated material may in certain circumstances be assessed by the criteria applied for expert opinion evidence (known as the Mohan criteria, after the Supreme Court case of R v. Mohan [1994] 2 SCR 9) to determine its admissibility in legal proceedings. By way of further example, the use of AI in violation of Charter-protected rights could also provide a basis for excluding evidence.

The Supreme Court of Canada in R v. Stinchcombe [1991] 3 S.C.R. 326 established the overarching constitutional requirement for the Crown to disclose all 'fruits of the investigation' and all relevant information to the defence. The Public Prosecution Service of Canada Deskbook sets out guidelines for federal prosecutors in the exercise of their prosecutorial discretion and their disclosure obligations, encompassing 'documentary and other evidence', 'identification evidence' (particulars of any procedures used outside court to identify the accused), and material generally relevant to the case-in-chief, while noting that 'additional disclosure' may be made at Crown counsel’s discretion, balancing the principle of fair and full disclosure.

Finally, the Canada Evidence Act (R.S.C. 1985), c. C-5 applies to criminal law cases, and includes provisions in relation to the introduction of digital evidence in criminal proceedings (section 31). These include requirements as to authentication of electronic documents, a presumption of integrity with respect to the recording or storage of such evidence, and notes that standards or procedures by which electronic documents are recorded or stored may also be presented when assessing the admissibility of an electronic document.

Deepfakes and synthetic media

As at June 2026, Canada does not have general regulation of deepfakes or AI-generated misinformation. Following the Protecting Victims Act (Bill C-16), which received royal assent in June 2026, an amendment has been proposed to the definition of 'intimate image' in section 162.1 of the Criminal Code, to include an electronically or mechanically made visual representation of an identifiable person, if the depiction is likely to be mistaken for a visual recording of that person. This amendment is likely to be relevant to criminal cases involving the creation or distribution of sexual deepfakes.

Data protection legislation

Existing data protection obligations under federal and provincial law also apply to the use of AI and apply, where applicable, to all actors in the criminal justice process.

Personal Information Protection and Electronic Documents Act (2000)

First, the Personal Information Protection and Electronic Documents Act 2000 (PIPEDA) at the federal level applies to private-sector organisations which use personal information in the course of commercial activities. PIPEDA is principles-based, with 10 ‘fair information principles’ including accountability (organisations are responsible for personal information under its control), consent (knowledge and consent is generally required for collection, use or disclosure of personal information), limited collection and safeguards (appropriate security protections for personal information must be adopted relative to the sensitivity of the information). As such, its provisions may apply broadly to AI tools which collect or use personal information. Three of Canada’s provinces (Alberta, British Columbia and Québec) have separate privacy laws applicable to in-province entities, which in many respects are similar to PIPEDA.

Following a public consultation, the Office of the Privacy Commissioner of Canada published proposals in November 2020 for legislative changes to PIPEDA to provide for a more robust privacy regulatory framework for AI. Although the Artificial Intelligence and Data Act (see above) was intended to provide a new governance framework for AI in place of PIPEDA, as at June 2026, that Bill has not been reintroduced by Parliament.

Protecting Privacy and Consumer Data Act (Bill C-36) (2026)

As at June 2026, Canadian privacy law is also pending reform through Bill C-36, the Protecting Privacy and Consumer Data Act, which would modernise private-sector privacy law and increase transparency around automated decision systems, including those powered by AI. If enacted, the Bill would repeal Part I of the Personal Information Protection and Electronic Documents Act. The Bill recognises the 'fundamental right of privacy of individuals with respect to their personal information'. However, whereas the Bill’s predecessor (Bill C-27, the Digital Charter Act, which was introduced in June 2022 but then stalled when Parliament prorogued in January 2025) included the Artificial Intelligence and Data Act (see above) in its Part 3, and would have created a framework for regulating high-impact AI systems, the new Bill C-36 does not include any equivalent. The Bill does, however, require organisations to disclose their use of automated decision systems that 'could have a legal or similarly significant effect' on individuals.

Privacy Commissioner of Canada Principles for Responsible, Trustworthy, and Privacy-Protective Generative AI Technologies (2023)

On 7 December 2023, the Office of the Privacy Commissioner of Canada published joint guidance with all Canadian provincial and territorial privacy regulators on Principles for Responsible, Trustworthy and Privacy-Protective Generative AI Technologies. These principles consider existing privacy laws as they apply to generative AI. They do not expressly prohibit the use of generative AI for administrative decision-making (whether fully automated or otherwise), but provide that any such use must be clearly communicated to affected parties, as well as explanations as to the general description of the system, how it is used to make a decision or take an action and an overview of potential outcomes. The principles also note that organisations using generative AI in such a manner remain accountable for decisions made, and that there should be an opportunity to request human review or reconsideration.

Provincial regulations

Provincial privacy regulators have also published specific guidance on AI systems where they arise. For instance, the Information and Privacy Commissioner of Ontario first published Guidance on the Use of Automated Licence Plate Recognition Systems by Police Services in July 2017. They then updated the Guidance in December 2024, noting that such systems pose risks to the privacy of individuals and that proper policies, procedures and technical controls must be adopted to ensure that personal information is handled lawfully.

Human rights

The Charter of Rights and Freedoms 1982 includes a number of provisions relevant to the use and deployment of AI in criminal proceedings:

There are also provincial human rights instruments and bodies that have addressed the intersection of human rights and AI at the provincial level. For example, the Ontario Human Rights Commission together with the Law Commission of Ontario has published a Human Rights AI Impact Assessment Tool, which is intended to enable organisations to assess their AI systems for compliance with human rights obligations.

Canada is also a party to international human rights instruments, including the International Covenant on Civil and Political Rights (Articles 14 and 17) and the Convention on the Rights of the Child (Articles 16 and 40) which enshrine similar guarantees in connection with fair trial rights and privacy.

Outlook

The Law Commission of Ontario is also developing a project on intimate images and deepfakes. The Commission’s project examines how a civil legal framework could address the creation, alteration, and distribution of intimate images without consent, including who should be held responsible, how liability should be assessed, and whether protection should depend on a reasonable expectation of privacy. However, as at June 2026, this project is in its early stages. 

CASES

Facial recognition

In 2021, a joint investigation by the federal Privacy Commissioner of Canada and the respective privacy commissioners of British Columbia, Alberta and Québec’s respective privacy commissioners examined the use of Clearview AI in Canada by Canadian law enforcement agencies and private organisations to identify individuals. They found that Clearview AI violated PIPEDA and provincial privacy legislation by 'scraping' images from internet websites without users’ consent, describing this as a form of 'continual mass surveillance' and a clear violation of Canadian privacy rights. The investigation noted that social media posts by individuals cannot be treated as falling under the 'publicly available' exception in PIPEDA, and that collection from these sources would need to be authorised by consent.

The findings and consequential orders were subject to challenge by way of judicial review in each of the three provincial jurisdictions:

• In British Columbia, in Clearview AI Inc. v. Information and Privacy Commissioner for British Columbia 2024 BCSC 2311, the Court upheld the British Columbia Commissioner’s order in its entirety. On 18 February 2026, the British Columbia Court of Appeal dismissed an appeal by Clearview and reaffirmed the British Columbia Supreme Court’s order: 2026 BCCA 6.
• In Alberta, in Clearview AI v. Alberta (Information and Privacy Commissioner) 2025 ABKB 287, the Court upheld the Alberta Commissioner’s conclusion that Clearview did not have a reasonable purpose for collecting, using and disclosing personal information. However, the Court also accepted aspects of Clearview’s Charter argument and held that parts of Alberta’s Personal Information Protection Act and its regulation unjustifiably limited freedom of expression insofar as they imposed a blanket consent requirement for the collection, use and disclosure of publicly available internet information. The Court otherwise maintained key parts of the Alberta Commissioner’s decision.
• As at June 2026, the decision on the challenge raised before the Courts of Québec (Clearview AI Inc. c. Commission d’accès à l’information du Québec) is still pending.

Sentencing

The dangers of the use of technological tools in the criminal justice context were highlighted in the Supreme Court of Canada’s judgment in Ewert v. Canada 2018 2 S.C.R. 165. In this case, the defendant was an indigenous offender serving concurrent life sentences for murder and attempted murder. He challenged the use of certain psychological and actuarial 'risk assessment tools' by the Correctional Service of Canada as breaching the relevant statutory requirements in the Corrections and Conditional Release Act as well as his rights under sections 7 (life, liberty and security of the person) and section 15 (equality) of the Charter of Rights and Freedoms.

One of the central findings of the Court’s judgment concerned accuracy of information. The Court was asked to determine whether the Correctional Service of Canada’s use of the tools without adequate research confirming their accuracy as it related to indigenous peoples was contrary to a requirement under the Corrections and Conditional Release Act to take ‘all reasonable steps’ in ensuring that the information it uses when making decisions about offenders was ‘as accurate, up to date and complete as possible’. In circumstances where the tools were ‘susceptible’ to cultural bias (as found by the trial judge), and the Correctional Service of Canada failed to conduct research to confirm the validity of the tools as applied to indigenous inmates, the Court found that the Correctional Service had breached that statutory duty. However, the Charter of Rights and Freedoms challenge, based on human rights law, failed. Ultimately, the Court considered it unnecessary to address those arguments given its statutory findings, leaving the possibility of such challenges open for future cases.

Electronic disclosure and the conduct of litigation

Canadian courts and regulators have recognised the need to adopt technological solutions with respect to disclosure and the wider conduct of litigation. 

For instance, the Ontario Superior Court in L’Abbé v. Allen-Vanguard 2011 ONSC 7575 (a civil case concerning allegations of breaches of contract and warranty, misrepresentation and fraud in connection with a sale purchase) has extolled the use of technology to ensure proportionality in disclosure, specifically referring to predictive coding and auditing procedures such as sampling. Master MacLeod in that case noted:

‘I accept that faced with this volume of documents, new approaches must be adopted but this cannot be a unilateral exercise. It requires ongoing procedural collaboration with court direction if necessary. Collaboration will not always result in agreement but where agreement is not possible, transparency should be the order of the day. Faced with this number of documents, the parties and the court must re-evaluate traditional approaches. Case law developed for manageable numbers of paper based documents must also be re-evaluated. Painstaking scrutiny of each individual document is disproportionate to the objective and unjustified even for a claim of this magnitude. Technology must be harnessed. Creative solutions need to be embraced. Counsel owe it to their clients and to the administration of justice to find efficiencies without, obviously, sacrificing the objective of a just outcome.’

In Drummond v. The Cadillac Fairview Corp. Ltd. 2018 ONSC 5350, an occupier’s liability case, the Ontario Superior Court when approving the recoverability of disbursements spent on legal research indicated that 'computer-assisted legal research is a necessity for the contemporary practice of law and computer assisted legal research is here to stay with further advances in AI to be anticipated and to be encouraged.'

In a further interlocutory decision of the Ontario Superior Court (Worsoff v. MTCC 1168 et al. 2021 ONSC 6493), Justice Myers made clear that 'Counsel and the court alike have a duty of technological competency', referring in that regard to the use of technology as a means of improving access to civil justice.

Misuse of AI in court filings

The misuse of AI by legal professionals and laypersons has featured in numerous criminal and non-criminal cases. Selected cases are discussed below.

Misuse of AI by lawyers

In Zhang v. Chen 2024 BCSC 285, a family case, counsel for the respondent cited two non-existent authorities in the notice of application. She admitted they were suggested by ChatGPT and filed an affidavit explaining the circumstances giving rise to her reliance on the hallucinated cases. In its judgment, the British

Columbia Superior Court pronounced that ‘[c]iting fake cases in court filings and other materials handed up to the court is an abuse of process and is tantamount to making a false statement to the court. Unchecked, it can lead to a miscarriage of justice.’ However, taking into account the circumstances, the Court declined to award special costs personally against the lawyer (which required reprehensible conduct or abuse or process), noting there had been no intention by the lawyer to deceive or misdirect and recognising the apology offered to counsel and the Court. But the Court did find that another provision of the Court rules enabled the Court to order that the lawyer be personally liable for all or part of the costs paid to another party, and in this circumstance, ordered that additional expenses incurred due to the insertion of fake cases be borne personally by the lawyer.

In contrast, costs were imposed by the Federal Court in Hussein v. Canada (Immigration, Refugees and Citizenship) 2025 FC 1060 in connection with the applicants’ underlying claim for refugee status which was refused and their order for removal from Canada. In Hussein, the applicants’ counsel was sanctioned for undeclared use of AI. The lawyer admitted to using a legal research platform called Visto.ai, and indicated he had not independently verified the citations. He also contended that his argument was unaffected and supported by other cases. The Court explained that this was not permissible. In its judgment, the Court indicated that ‘[t]he use of generative artificial intelligence is increasingly common and a perfectly valid tool for counsel to use; however, in this Court, its use must be declared and as a matter of both practice, good sense and professionalism, its output must be verified by a human.’ The Court found that counsel’s approach in response to the Court’s directions was an attempt to mislead the Court and ordered there were special reasons warranting an award of costs as a result.

Ko v. Li resulted in two separate decisions. In 2025 ONSC 2766, the Ontario Superior Court found there were numerous fabricated case citations in the applicant’s counsel’s factum. He identified various professional duties, including that it was ‘the lawyer’s duty to ensure human review of materials prepared by non-human technology such as generative artificial intelligence' and that 'it should go without saying that it is the lawyer’s duty to read cases before submitting them to a court as precedential authorities’. The judge directed a show cause hearing for the applicant’s lawyer as to why she should not be cited for contempt of court. The subsequent contempt proceedings (2025 ONSC 2965) resulted in a dismissal of those contempt proceedings taking into account the relevant facts. The Court indicated that the lawyer had purged any possible contempt, given that she had admitted the relevant facts as to her use of AI (ChatGPT) in the preparation of her factum, apologised, and had undertaken to take courses in the proper use and risks of AI in practice.

As already discussed above, R. v Chand 2025 ONCJ 282 involved the use of AI by a criminal defence lawyer; in that case, the Court did not impose sanctions but instead directed that counsel, who had otherwise 'done a good job presenting the defence in this case', resubmit his submissions after having done a full cite-check and without the use of generative AI or legal research using generative AI.

Misuse of AI by self-represented litigants

Use of AI by a self-represented litigant was considered in Specter Aviation Limited c. Laprade 2025 QCCS 3521, a case concerning the ratification of an arbitration award. The litigant in that case used AI tools to prepare his submissions, giving rise to case hallucinations. The Superior Court of Quebec indicated that this was the first instance in which it was asked to consider the issue of hallucinated citations from AI and suggested that the issue was likely to ‘fill many pages of case law’ in the future. It found that there had been a significant breach of procedure and fined the litigant C$5,000.

In another case involving a self-represented litigant, in Lloyd’s Register Canada Ltd. v. Choi 2025 FC 1233, a motion record by the litigant included a non-existent case citation to support a proposition about the Court’s discretion to grant a subpoena. The Court made reference to the Court’s AI Practice Direction, which required disclosure to the court of the use of AI, and in this case, the litigant continued ‘to insist he has done nothing wrong.’ Accordingly, the Court removed the motion, indicating doing so was 'necessary to preserve the integrity of the Court’s process and the administration of justice' and awarded costs.

Deepfakes and synthetic media

There are an increasing number of criminal cases in the Canadian courts which address the issue of deepfakes.

R. c. Larouche 2023 QCCQ 1853 was the first case in Canada to deal with the use of deepfakes (‘hypertrucage’ in French) with respect to the offence of making child pornography. The case concerned the sentencing of a man who was charged and pleaded guilty to child pornography offences. Among some of the files retrieved were over 86,000 child pornography files generated using deepfake technology. The total volume of information retrieved was said to be ‘among one of the most extensive in the Court’s history.’

As the Court explained, deepfake technology ‘is an audiovisual technique that uses deep learning algorithms to create extremely realistic fakes.’ In this case, the Court was considering software which it explained drew on a bank of ‘source’ and ‘destination’ material (e.g. photographs and video clips). The software ‘sequences a video excerpt image by image to obtain a bank sufficient to create a minimally realistic deepfake. Once the database is sufficiently complete, the software tries to teach the artificial intelligence to take into account the different facial features on each photograph: angle of the face, position of the eyes, lips, ears, etc. to mimic the source face’s movements.’

The Court warned about the particularly acute social impact the use of deepfake technology could have:

‘The use of deepfake technology by criminals is chilling. This type of software makes it possible to commit crimes that could involve practically every child in our communities. A simple video excerpt of a child on social media or a surreptitious screenshot of children in a public place could transform them into potential child pornography victims. All a cybercriminal has to do is sequence the video and exchange the child’s face with the face of a sexual assault victim found online. New files are thus created and the children’s sexual and psychological image and integrity are irreparably harmed, with the potential for that file to be disseminated everywhere on the Internet, without any control.’

In doing so, it also recognised that ‘[t]he technology currently used by law enforcement is now ineffective and will rapidly become obsolete.’

In relation to the sentencing aspect of the decision, the Court was asked to determine the appropriate sentence for the offence of making child pornography, noting that in the case of the production of deepfakes, there were well over 86,000 files made (and multiplied due to between 15-30 individual photo files being combined to create each second of video). In its assessment, however, the Court indicated that using deepfake technology does not necessarily warrant exceptional treatment in terms of assessing sentencing, since ‘using technology to make deepfake material does not substantially change the essence of the offence of making child pornography.’ Although the Crown sought a sentence for this aspect of 6 years’ (72 months) detention, the Court considered, taking into account all the factors (including the lack of evidence as to distribution, the guilty plea and cooperation with the police, among others), that the appropriate sentence was 42 months’ detention.

Detection and assessment of deepfakes are also a particular concern for courts.

For instance, in R. v. Analib-Goortani 2014 ONSC 4690, the Ontario Superior Court was asked to consider the authenticity and ultimately, admissibility, of a photograph said to have been taken during a protest at the G20 summit held in Toronto, in which a police officer (the accused) was depicted as about to hit a female with a baton. The accused was charged with assault with a weapon. Both the Crown and defendant’s counsel adduced expert evidence to testify about the provenance of the image and whether it may have been altered. It was accepted that the image, which had been sourced from a website, had been largely stripped of its metadata. The Court ultimately determined that the photograph had not been properly authenticated and so was not admissible. In its judgment, the Court indicated that ‘[i]n a criminal trial, it would be improper for counsel to wave a photograph around in front of the jury, or thrust it under the nose of a witness, without first addressing the issue of authenticity. The potential for unfairness is obvious.’ It also warned that ‘[m]aterial taken from websites and offered as evidence in court must be approached with caution, especially in a case such as this where no one is prepared to step forward to say, ‘I took that photo and it has not been altered or changed in anyway.’ While the Court was not directly being asked to find that this particular photograph was a ‘deepfake’, the issue of authenticity it was grappling with is one which has direct bearing on the (mis)use of deepfake evidence in criminal trials.

The need for credible expert evidence to assess deepfakes presented to the Court is likely to continue to be an issue going forward. In the case of R. c. Azubuike 2024 QCCS 1654, one of the issues before the Superior Court of Quebec in this fraud case was whether certain phone call recordings being relied on in evidence by the Crown against the accused may have been 'cloned through Artificial Intelligence'. The defence’s nominated witness was not, per the Court, ‘a qualified expert witness in the field of Artificial Intelligence specializing in voice cloning and voice analysis’; implicit in that finding is the recognition that this is an area where a court would be assisted by a duly and properly qualified expert. The accused was eventually found guilty by a jury (2024 QCCS 1976).

By contrast, in R. v. MD 2024 ABKB 104, a sexual assault case before the Court of King’s Bench of Alberta, the defence contended that a recording of a conversation relied on by the Crown may have been created by AI. This was dismissed by the judge as ‘entirely speculative’ noting there had been no evidence ‘that SD [the victim, who had taken the recording using her cellphone] had any familiarity with AI or other technology that might be used to create such a dialogue, that she knew any persons who did, that SD had ever done anything like that before, that it was in fact possible, by whatever technique, to create a dialogue such as that (apparently) recorded, that she had any particular motive to so fabricate, or that forensic analysis of the recording confirmed or even suggested that the dialogue was fabricated, even in part’. The Court ruled and made findings in relation to the recording, considering it in light of the entire evidence, and found the defendant guilty of the charged offences.

Further, in R. v. Medow 2025 ONCJ 661, the accused was charged with numerous offences, including obstruction and assault. The Ontario Court of Justice was asked to determine whether video evidence from body-worn camera footage and in-car camera footage ‘might be a ‘deepfake. In that case the Court accepted:

‘that the existence and widespread proliferation of AI technology capable of producing realistic deepfake videos are matters properly subject to judicial notice: see R. v. Cheng, 2025 ONCJ 252, at para. 7. Anyone who spends time online has likely encountered well-known deepfake videos, particularly those featuring celebrities or politicians. These can be highly deceptive, making it challenging to discern what is authentic and what is fiction. Applications that help users create these videos are heavily marketed and promoted on various social media platforms, including Twitter/X and TikTok. Many more can be purchased and downloaded from various websites. Recently, the Supreme Court of Canada noted that the production of deepfakes through artificial intelligence for criminal purposes, such as the creation of child sexual abuse and exploitation material, is a 'present and growing danger': see Quebec (Attorney General) v. Senneville, 2025 SCC 33, at para. 32.’

However, in this case, the Court noted the accused had not presented any evidence suggesting the videos had been manipulated apart from the blurring of a motorist’s face and found that ‘mere speculation by an accused person that digital evidence has been falsified to deceive the viewer is insufficient to preclude its admissibility or to diminish its weight.’ The judge, however, was keen to foreshadow that:

‘[a]lthough I have ruled that the officers' testimony sufficiently authenticated the videos in this case, Justice Nordheimer’s warning in Aslmani that some forms of digital evidence may require the Crown to present expert evidence for authentication should be taken very seriously.

[…]

Courts must seriously consider how to assist an economically disadvantaged self-represented accused person who disputes the authenticity of digital evidence to ensure a fair trial, without compromising their independence. While there are ongoing efforts in the scientific community to develop tools for courts to identify deepfakes, they have not yet come to fruition, to the best of my knowledge.

[…]

As generative AI technology continues to advance and the ease with which viewers can be deceived by falsified digital evidence increases, courts must ensure that the authentication voir dire required for digital evidence is not rendered meaningless. This presents a potentially serious access-to-justice issue, as the vast majority of self-represented litigants, such as Mr. Medow, will lack the means to retain an expert of their own. Nothing in this ruling should be interpreted as a decision that the Crown is never required to present more comprehensive evidence than was done in this trial to ensure that its digital evidence can be safely relied upon. Each case must be assessed independently, and the law may soon need to adapt given the stark realities of these ever-changing technologies and their capacity to negatively impact on the truth-seeking function of the criminal trial process’.