Protecting AI Innovation: Fiduciary Duties in the Age of AI

For companies building artificial intelligence (“AI”), it is essential to treat your software as one of your company’s most valuable assets and have safeguards in place to act quickly in the event of any misuse. Protecting AI assets isn’t just smart business; it’s a legal necessity that will preserve enforceable rights, avoid costly disputes and promote smooth commercialization.

The decision of the Ontario Superior Court of Justice in Arc Compute v. Anton Allen, Michael Buchel et al, 2025 ONSC 1745, serves as a reminder  that even in the absence of an AI legal framework, Canadian Courts are equipped to protect against AI-era intellectual property (“IP”) theft. The case involved former executives accused of misappropriating proprietary AI models and training data to launch a venture competing with their former employer.

The plaintiff, Arc Compute (“Arc”), is in the business of proprietary AI-models, associated training data sets, software and source code. In the action, Arc sought, and succeeded in achieving, an interlocutory injunction to restrain the defendants from retaining and disclosing confidential information belonging to Arc when the defendants started a new venture with Arc’s proprietary information. Notably, this decision demonstrates how AI-related IP, in the training data and model architecture spheres, can be protected under existing legal frameworks.

Background

One of the defendants, Michael Buchel (“Buchel”), joined Arc in 2021, as a Director of Engineering responsible for leading all aspects of Arc’s software business. The other defendant, Anton Allen (“Allen”), joined Arc in 2022, as Vice President of Sales. In both their employment agreements, Buchel and Allen agreed that Arc would own all the IP, ideas, research and inventions made by them during their employment with Arc; that they would not solicit Arc’s clients or employees; and that they would keep Arc’s confidential information confidential per their employment agreements.

In 2024, Arc experienced serious financial difficulties and terminated most of its workforce, including Allen, and subsequently, Buchel. Just before their respective terminations, Allen and Buchel established a new company and contacted Arc’s clients, using Arc’s proprietary information, and prepared to offer services that competed with Arc’s business using the very AI model Buchel had develop for Arc. Buchel retained a copy Arc’s source code, a direct violation of his employment contract.

The Court’s Finding: Fiduciary Duty Meets AI Misuse

The issues before the Court addressing the misuse of proprietary AI IP were as follows:

1. Fiduciary duty and the misuse of confidential information – Did Allen and Buchel owe a fiduciary duty to Arc?
2. Non-competition and the protection of the company’s AI assets – Did Allen and Buchel utilize Arc’s confidential data (including source code, software and training data sets) to launch a competing venture?

Although the Court’s discussion focuses on Canada’s current lack of AI legal governance regime, this case signals that the existing common law concepts of fiduciary duties, confidentiality and contractual enforceability can sufficiently protect AI assets.

The Court assessed Buchel and Allen’s actions in light of the corporate opportunity doctrine. The corporate opportunity doctrine precludes fiduciaries from obtaining for themselves, any property belonging to the business, and restricts fiduciaries from diverting to themselves opportunities that their company is pursuing. Both Buchel and Allen were found to be fiduciaries of Arc via the common law test as articulated in Frame v. Smith, which found that an employee’s fiduciary duty continues after their employment ceases. 

The Court ordered the defendants to shut down their new venture, on the basis that it is a corporate opportunity belonging to Arc. Further, Buchel and Allen were restrained from competing with Arc given that the software was a brand new product still in its experimental phase.

Key Takeaways

This judgment highlights an increased need for robust AI governance around innovation, especially in competitive commercial environments. Companies with clear employment contracts assigning IP rights, confidentiality obligations, exit protocols and data control measures can protect proprietary AI IP via established legal frameworks.

In these high-stakes environments, where the misappropriation of source code could cause serious competitive harm, Courts will permit injunctions to avoid irreparable damage. However, it is critical that companies treat the software behind its AI technology as a core corporate asset and put adequate safeguards in place in employment agreements to allow for swift enforcement against any misuse by rogue employees. Protecting AI assets isn’t just prudent; it’s legally critical to preserve enforceable rights and prevent costly disputes.

For more information about the legal implications of the use of AI technology, please contact Roland Hung and Laura Crimi of Torkin Manes’ Technology and Privacy & Data Management Groups.

 *The authors would like to thank Torkin Manes’ Articling Student Ilar Haydarian for her invaluable contributions in preparing this insight.