By: Suzana Popovic-Montag
The appeal of using artificial intelligence (“AI”) to generate documents is undeniable. AI is convenient, efficient and cost-effective – professional appearing documents can be generated in minutes, if not seconds, upon receiving input data, and at a fraction of the price it would typically cost to have a document drafted by a person. Legal documents, including wills, can even be generated using these platforms.[1]
Despite the advantages of using generative AI to create documents, there can also be negative consequences. For example, in Ko v Li,[2]a case recently discussed on our firm blog, counsel utilized generative AI to prepare court submissions which turned out to feature “AI hallucinations,” citing inaccurate and fictitious caselaw.[3] This case demonstrates the danger of relying on generative AI – while the documents these platforms create may appear to be professional, correct, and reliable, they may in fact be erroneous and misleading.
It is probable, if not inevitable, that wills and estates practitioners will one day have to deal with the fallout of testators using generative AI platforms to create wills. This article contemplates a specific issue related to probate – whether an AI generated will can be rectified on the basis that the AI program made a drafting error.[4] It is unclear at this time whether such wills could be rectified beyond correcting simple typographical errors – it may be difficult to prove that an AI program misunderstood a testator’s will instructions, or, depending on the quality of instructions, that they were not carried out.
The Risk of AI-Generated Wills Containing Mistakes and Drafting Errors
In keeping with the submissions drafted by AI in Ko v Li,[5] there is a risk that wills created by generative AI, subsequently referred to as “AI-wills,” may contain drafting errors that would result in the deceased’s estate being distributed in a manner contrary to their intentions. This risk seems particularly plausible given that generative AI programs do not actually understand the content that they create. Generative AI harnesses machine learning rather than true “intelligence,”[6] and uses large language models to learn the patterns and structure of its input training data, which is then used to generate new data with similar characteristics.[7] In other words, generative AI “uses complex machine learning models to predict the next word based on previous word sequences.”[8]
By creating a will this way, without genuinely understanding either the user’s instructions or the uploaded wills upon which the AI-will is premised, there is an obvious risk that the output could repeat or otherwise introduce mistakes or drafting errors, especially depending on how the wills that the AI-will is modelled on were drafted.[9] Potential drafting issues that could arise, to name a few, include the incorrect use of legal terminology, ambiguous will clauses, contradictory will clauses, the creation of gifts personae designatae rather than class gifts or vice versa, or the creation of wills that are missing essential will clauses.[10]
If a layperson is working with generative AI to create a will and is presented with what appears to be a bona fide professional instrument, it may be quite difficult for them to catch drafting errors. It is also improbable that the AI program would bring potential issues or errors to the user’s attention.[11] Generative AI is notorious for including fabricated information in its outputs which appears authentic, and for presenting those outputs in a confident and compelling manner.[12] Moreover, if the document generated by AI looks like a professionally drafted will, garnering the confidence of the client, drafting errors may be overlooked, particularly if the user believes that a will ought to look like the document that was generated.[13] As with lawyer-prepared wills, a testator may not carefully review a will generated by AI, simply signing it after a cursory review of the document.
The Power to Rectify Wills Containing Drafting Errors
If an AI-will includes mistakes that come to light after the testator has passed away, the estate trustee could apply to the court to rectify the will and correct those errors. Rectification is an equitable doctrine in Ontario;[14] sitting as a court of probate,[15] the Superior Court of Justice can vary the terms of a will, usually by deleting words,[16] so long as the testator’s instructions were recorded incorrectly by mistake. However, this doctrine can only be utilized if the mistake was made by the will drafter, and it can be demonstrated that the testator did not approve of the words used by virtue of their instructions.[17]
A will cannot be rectified because of a misunderstanding on the part of the testator regarding the application or effect of the law, or because it produces an undesirable or unintended outcome. The presumption of knowledge and approval applies to a will even if a testator did not understand its legal implications, or had a mistaken belief about the legal effect of the words used.[18] Similarly, if a mistake is made by a testator who drafts their own will, the court may find that the presumption of knowledge and approval applies unless it is possible to establish that the will was “made in ignorance of some pertinent information.”[19]
Rectification is typically available only in three situations:
- where there is an accidental slip or omission because of a typographical error or clerical error;
- where the testator’s instructions have been misunderstood; or
- where the testator’s instructions have not been carried out.[20]
Before a court will rectify a will, the drafting lawyer is expected to provide evidence confirming the drafting error,[21] although rectification may also be granted based on evidence of surrounding circumstances.[22] The court may even hear direct evidence of the testator’s intentions.[23] Regardless of the nature of the evidence submitted, rectification can only be granted if that evidence has a “high degree of clarity, persuasiveness and cogency.”[24] In the context of estates, for example, rectification may be granted if the evidence establishes that the testator’s instructions were clear, and the drafting lawyer admits that they did not carry out the testator’s instructions in drafting the will.[25]
Assessing the Potential Rectification of AI-Wills
To date, it appears that no one has approached the court seeking rectification of an AI-generated will. There are a number of reasons why a court may be reluctant to grant this relief, however. If there is no evidence establishing who prepared the AI-will, for example, or if the testator’s “instructions” were not clear, or there is doubt as to whether the testator’s instructions were “misunderstood” by the AI-program or not carried out, an application for rectification may be denied. Moreover, given that rectification is typically granted for wills drafted by lawyers who have been retained to prepare a will in accordance with the testator’s instructions, it is feasible that a court would decline to extend the doctrine to AI-wills since AI cannot be retained as such. In fact, if the AI platform warned the testator that it was not qualified to give legal advice or draft legal documents, the court may decline to grant relief.
The Challenge of Rectifying an AI-Will if the Will Drafter is Unknown
In Canada, documents generated by AI may be mistaken for documents drafted by professionals, given that there is no requirement to mark documents which are a product of generative AI.[26] Therefore, an AI-will could be admitted for probate without the estate trustee or the court even knowing that it is an AI-will. If rectification is sought but the identity of the will drafter is unknown, undermining the applicant’s ability to establish the testator’s actual instructions, a court may be reluctant to rectify a will beyond correcting obvious typographical errors.[27] As previously noted, a will cannot be rectified without evidence which has a “high degree of clarity, persuasiveness and cogency.”[28]
Admittedly, it may be possible to obtain rectification if other sources, such as the testator’s friends and family, can provide evidence of the testator’s instructions. However, as noted by Justice Belobaba, the court is often “less comfortable relying on affidavits (often self-serving) from putative beneficiaries who purport to know what the testator truly intended.”[29]
The Challenge of Establishing the Testator’s Instructions Through AI Prompts
If the estate trustee is aware that the will was generated by AI, putting evidence of the testator’s “instructions” before the court in order to seek rectification may be challenging. To accomplish this feat, the applicant would likely need to obtain the prompts provided to the testator by the AI program and the answers provided by the testator. This information may simply be unavailable, depending on whether it was saved by the testator or can still be found in the testator’s history for the generative AI program. The applicant also may be unable to access this information, even if it exists.
That said, if the AI prompts and the testator’s answers can be located, the applicant could put it before the court through an affidavit, attached as an exhibit. If the testator gave their instructions to the AI program in the company of a third party, that party could swear the affidavit; otherwise, the applicant could. Unfortunately, such evidence may not be as persuasive or cogent as that typically put forward on a rectification application from a drafting lawyer. Evidence from the actual will drafter is preferable because the will drafter can confirm both (1) the testator’s instructions and (2) that a drafting error was made. A generative AI program simply cannot provide such evidence.
The Challenge of Establishing that AI “Misunderstood” the Testator’s Instructions
It may also be a challenge for the applicant to prove that the testator’s instructions were misunderstood by the will drafter when seeking rectification of an AI-will. Technically speaking, AI is unable to truly understand – or misunderstand – instructions, since generative AI utilizes machine learning rather than true intelligence. It also seems improbable that anyone who works with the AI program would have been involved in interpreting the testator’s instructions, given that such programs utilize unsupervised or semi-supervised machine learning algorithms.[30] Because of this technicality, rectification may be unavailable for an AI-will on the basis of the testator’s instructions being misunderstood.
The Challenge of Establishing That the Testator’s Instructions Were Not Carried Out
When probating an AI-will, a court likely could infer that the testator’s instructions were not carried out by the AI program. However, the ability to make such an inference would likely turn on the quality of prompts provided by the AI program, the quality and clarity of the testator’s “instructions,” and the amount of information provided by the testator to the AI program. If the information provided by the testator is minimal or unclear, it may not be feasible to establish that those “instructions” were not carried out.
AI Is Not Qualified to Provide Legal Services
The law also has not yet addressed whether rectification should be available for AI-wills. It could be argued that this relief should not be extended as such, given that AI is not qualified to draft legal documents and makes no actual representation to an AI user that it is capable of drafting a valid will using information provided by the user.
Typically, rectification is granted for wills drafted by lawyers who failed to carry out the testators’ instructions. The law recognizes that if a lawyer is retained to draft a will, the testator may trust that the will prepared by the lawyer reflects their instructions. As noted in Parker v Felgate, upon giving instructions to a lawyer to make a will and reviewing the lawyer’s work, the client should be able to think: “I gave my solicitor instructions to prepare a will making a certain disposition of my property; I have no doubt that he has given effect to my intention, and I accept the document which is put before me as carrying it out.”[31] If the lawyer drafts a will that is inconsistent with the client’s instructions, the client may be completely unaware, recognizing that a testator may have “no intelligent appreciation of the effect of [an] expression used by [a] solicitor.”[32] Given the trust that a testator puts in their lawyer to carry out their wishes, merited by virtue of a lawyer’s training and professional experience, it makes sense that rectification would be available if a will drafted by a lawyer includes a mistake.
When a will is created by a generative AI program, however, there is no reasonable basis for the testator to trust that the program will follow their instructions when generating a will. General generative AI platforms are not trained as lawyers and are not authorized to provide legal services. In fact, an AI program may even warn its users that it cannot provide legal advice before generating an AI-will, in which case it could be argued that the testator received notice that an AI-will may produce undesirable or unintended outcomes, and chose to accept that risk. Under such circumstances, rectification may not be available.
On the flip side, if it is clear that an AI-will contains a drafting error, it could be argued that it would be unreasonably punitive to the testator to withhold rectification on the basis that their will was not drafted by a lawyer, or that it would be draconian to fail to extend such equitable relief to modern methods of will drafting. As held by Justice Sachs, “it is not the law of this country that the testator can give testamentary validity to a testamentary disposition by accepting without understanding its effect something put forward by another.”[33]
Conclusion
Given the appeal of generating documents with AI, it is only a matter of time before AI-wills are admitted to probate. When this occurs, one issue that may arise is the availability of rectification to correct drafting errors in AI-wills. As explained in this article, it may be difficult to obtain rectification for mistakes beyond basic typographical errors, as it will be challenging (if not impossible) to meet the current test for rectification with an AI-will. The outcome of an application will likely turn on whether it is known that the will was generated by AI, whether the testator’s instructions, provided in response to AI prompts, are clear, the amount of information provided by the testator, the quality of prompts provided by the AI program, and whether the court can infer that the testator’s instructions were not carried out. There is also an overarching question to consider – recognizing that AI is not qualified to provide legal services, does the testator implicitly accept the risk that an AI-will may produce undesirable or unintended outcomes? And, if so, does this mean that rectification is no longer available? Recognizing that there are no easy answers to these questions, it will be interesting to see how mistakes in wills generated by AI are handled by the courts in the future.
Editor: Suzana Popovic-Montag
141 Adelaide St. W. Suite 1700 Toronto, Ontario
M5H 3L5 Tel: 416-369-1140 spopovic@hullandhull.com
Please note that all back issues of The Probater are available in full text on our website, www.hullandhull.com
[1] See James Peacock, “An AI Wrote My Will. I’m an Estate Lawyer. Goodbye Career.” (17 April 2023), online: After Your Time <https://afteryourtime.com/chatgpt-ai-generated-will/>. Also see Tolou Mahani, “The impact of ChatGPT on estate planning” (28 March 2023), online: Law 360 Canada <https://www.law360.ca/articles/45248/the-impact-of-chatgpt-on-estate-planning-tolou- mahani?article_related_content=1> [Mahani].
[2] 2025 ONSC 2766. In terms of sanction, see Ko v Li, 2025 ONSC 2965.
[3] See Mark Lahn, “AI Hallucinations in the Courtroom: Lessons from Ko v. Li” (20 May 2025) Hull & Hull LLP, online (blog): <https://hullandhull.com/2025/05/ai-hallucinations-in-the-courtroom-lessons-from-ko-v-li/>. See also Shawnee Matinnia, “AI Hallucinations in the Courtroom Part 2: Recent Developments in Ko v. Li” (26 May 2025) Hull & Hull LLP, online (blog): <https://hullandhull.com/2025/05/ai-hallucinations-in-the-courtroom-part-2-recent-developments-in-ko-v-li/>.
[4] Another issue that could arise related to the probate of AI generated wills is whether such wills may be validated if they are not executed in compliance with the legislated formalities under the Succession Law Reform Act, RSO 1990, c S.26. To learn more about this issue, see Suzana Popovic-Montag, “Contemplating Probate of Wills Generated by Artificial Intelligence” (Nov 2023) Toronto Law Journal, online: Toronto Lawyers’ Association <https://www.tlaonline.ca/?pg=TorontoLawJournalPub&pubAction=downloadIssue&pubIssueID=46416>.
[5] Supra note 2.
[6] Canadian Centre for Cyber Security, “Generative artificial intelligence (AI) – ITSAP.00.041”, Awareness Series (July 2023), online: Government of Canada <https://www.cyber.gc.ca/en/guidance/generative-artificial-intelligence-ai-itsap00041> [Cyber Security].
[7] Law Society of Ontario, “Licensee use of generative artificial intelligence”, (April 2024) White Paper, online: LSO Technology Resource Center < https://lawsocietyontario-dwd0dscmayfwh7bj.a01.azurefd.net/media/lso/media/lawyers/practice-supports-resources/white-paper-on-licensee-use-of-generative-artificial-intelligence-en.pdf> at 4 [White Paper].
[8] Thomas H. Davenport & Nitin Mittal, “How Generative AI is Changing Creative Work” (14 Nov 2022), online: Harvard Business Review <https://hbr.org/2022/11/how-generative-ai-is-changing-creative-work>.
[9] The quality of any testamentary instrument that an AI program generates will surely hinge on the quality of the documents that have already been uploaded to the program. Generative AI often gathers information from the internet and utilizes unsupervised or semi-supervised machine learning algorithms: see Altexsoft, “Generative AI Models Explained” (4 Sept 2024), online: altexsoft <https://www.altexsoft.com/blog/generative-ai/> [Altexsoft]. With this in mind, it is in the realm of possibility that an AI-will may be modelled on other AI-wills, as compared to wills created by legal professionals. If this were to occur, any drafting problems in uploaded AI-wills could end up being compounded, impacting the quality of subsequent AI-wills.
[10] For example, in Tolou Mahani’s article, supra note 1, she describes asking ChatGPT to provide a will template that is valid in Ontario, and receiving a one-page document in return, “[e]ven though the most basic simple will has at least several pages of clauses that are recommended.”
[11] If a number of wills uploaded to the program are drafted differently, for example, it seems that the program would simply create a will that is “statistically the best response to [the user’s] prompt” rather than bring different drafting options to the user’s attention. The quoted text is taken from Cyber Security, supra note 6.
[12] See the White Paper, supra note 7 at 6.
[13] Clients may face the same difficulty when executing wills prepared by lawyers. See Albert H. Oosterhoff et al., Oosterhoff on Wills, 9th ed (Toronto: Thomson Reuters Canada Ltd, 2021)at 272, discussing Rhodes v. Rhodes (1882), 7 AC 192 (P.C.).
[14] Robinson Estate v. Robinson, 2010 CarswellOnt 4576, [2010] O.J. No. 2771 at para 25, aff’d 2011 ONCA 493, leave to appeal refused [2011] SCCA No 536 [Robinson]. See also Ihnatowych Estate v Ihnatowych, 2024 ONCA 142 at para 35 [Ihnatowych]. In other provinces, in comparison, legislative provisions have been enacted which authorize the courts to rectify wills. In British Columbia, see the Wills, Estates and Succession Act, SBC 2009, c 13, s 59. In Alberta, see the Wills and Succession Act, SA 2010, c W-12.2, s 39.
[15] Rectification may also be granted by a court sitting as a court of construction. See Ian M. Hull & Suzana Popovic-Montag, Feeney’s Canadian Law of Wills, 4th ed (Toronto: LexisNexis, 2000) (loose-leaf) at §§ 3.37, 12.6 [Feeney’s].
[16] Oosterhoff notes that courts of probate have the “power to strike out passages inserted by mistake, but … does not have the power to substitute words that the testator intended to use”: Oosterhoff on Wills, supra note 13 at 263. See also Connor Estate v. Worthing, 2020 BCSC 150 at paras 46 and 50, aff’d 2021 BCCA 231. However, the court also recognized in this case that there is a narrow exception under which a court may rectify a will by adding words: see para 56. Words have also been added to a will via rectification in Ontario: see Ihnatowych, supra note 14, aff’ing Gorgi v Ihnatowych, 2023 ONSC 1803.
[17] Feeney’s, supra note 15at § 3.28.
[18] Robisnson ONSC, supra note 14 atpara 30. Also see Feeney’s, ibid at § 3.30.
[19] See Feeney’s, ibid at § 3.30, citing Re Swords, [1952] P. 368, [1952] 2 All E.R. 281, a case where the testator made a codicil to her will herself, but without looking at the will; the court struck out certain words because it could not be said she approved of the codicil, having made it in ignorance of the wording of the will.
[20] See Ihnatowych, supra note 14, citing Robinson ONSC, supra note 14 at paras 24-25.
[21] See Robinson ONSC, ibid at paras 25-26; Justice Belobaba observed: “Courts are more comfortable admitting and considering extrinsic evidence of testator intention when it comes from the solicitor who drafted the will, made the error and can swear directly about the testator’s instructions.” See also Mansour v. Girgis, 2024 ONSC 1611at para 26.
[22] See Verity (Re), 2012 BCSC 650. In this case, the drafting lawyer had passed away and the assistant who helped prepare the testator’s will had moved to another firm.
[23] See Oosterhoff on Wills, supra note 13 at 267, 270.
[24] Ihnatowych, supra note 14at para 38, citing Canada (AG) v Fairmont Hotels Inc, 2016 SCC 56 at para 36 [Fairmont].
[25] See Ihnatowych, ibid at paras 39-40.
[26] While the federal government tabled the Artificial Intelligence and Data Act in 2022 as part of Bill C-27, the Digital Charter Implementation Act, 2022, Bill C-27 died earlier this year when Parliament was prorogued. See Blair Attard-Frost, “The Death of Canada’s Artificial Intelligence and Data Act: What Happened, and What’s Next for AI Regulation in Canada?” (17 January 2025), online: Montreal AI Ethics Institute <https://montrealethics.ai/the-death-of-canadas-artificial-intelligence-and-data-act-what-happened-and-whats-next-for-ai-regulation-in-canada/>.
[27] A court may correct a typographical error in a will without evidence of the testator’s instructions using the doctrine of falsa demonstratio. See, for example, Das Estate (Re),2012 NSSC 411 at paras. 26-29.
[28] Ihnatowych, supra note 14at para 38, citing Fairmont, supra note 24at para 36.
[29] Robinson ONSC, supra note 14at para. 26.
[30] See Altexsoft, supra note 9.
[31] Parker v Felgate (1883), 8 PD 171 (Eng PDA). This passage has been endorsed in Canada: see Faulkner v. Faulkner, 1920 CanLII 4 (SCC). See also McLaughlin (Estate of) v. McLaughlin, 2015 ONSC 4230 at paras 82-84.
[32] See Oosterhoff on Wills, supra note 13 at 272, citing Rhodes v. Rhodes (1882), 7 AC 192 (P.C.). It has also been noted that testators typically do “not understand the minutiae of a will, and cannot reasonably be expected to identify all errors within a will”: see WEL Partners, “Rectification of Wills: Drafting Error or Mistake by the Testator?” (29 September 2017), online: WEL Blog <https://welpartners.com/blog/2017/09/rectification-of-wills-drafting-error-or-mistake-by-the-testator/#_ednref8>.
[33] Crerar v Crerar, unreported, 1956, Ch., quoted in Lloyds Bank Ltd. v. Peake (1969), [1970] 1 All ER 1057, [1971] P 62 (Eng. P.D.A.) at 81 [P]. This passage is cited in Ian M. Hull & Suzana Popovic-Montag, Probate Practice, 6th ed (Toronto: Thomson Reuters Canada Ltd, 2023) at § 2:15.