The proliferation of AI tools that can generate convincing voice, image and video impersonations has led to increased (and often hidden) liability risk for companies leveraging AI technologies in their marketing.
In Canada, the misuse of one’s image is protected at common law and can be enforced by asserting either privacy rights or personality rights, depending on the situation.
Privacy rights
Focusing first on privacy rights: where a person is not famous and enjoys a reasonable expectation of privacy, the publication of that person’s likeness or personal information without permission may constitute an invasion of their right to privacy.
To establish a breach of one’s right to privacy, a plaintiff must establish that:
- the defendant’s conduct is intentional, or reckless;
- the defendant has invaded, without lawful justification, the plaintiff’s private affairs or concerns; and
- a reasonable person would regard the invasion as highly offensive causing distress, humiliation or anguish.
Marketers are thus well advised to obtain a release to feature a person in any marketing materials, and that the release corresponds to the actual use. For example, a release to be included in a short video may not be sufficient to later feature that person’s face on a billboard. The tort of breach of privacy was helpfully discussed in Vanderveen v Waterbridge Media Inc, a case in which a woman obtained damages after video of her jogging on a public, riverside path was featured for two seconds within a two-minute promotional video for a real estate development in the area. The plaintiff was disturbed to discover herself in the promotional video, and requested its removal, which occurred within one week. She then sued for damages. The Court found that the defendant’s including her in its promotional video was intentional or reckless, that she had not consented to being filmed, and that a reasonable person would regard the invasion as highly offensive.
So, when it comes to AI-produced images for commercial purposes, brand owners should be aware of the possibility that the AI may produce an image of an actual person. To the extent that the AI produces a mock-up or “imaginary” person, the risk is minimized. However, if the AI has been trained with real photographs and video and somehow includes a real person’s likeness in the AI-generated marketing material, there is a material risk of liability for breach of privacy.
Personality rights
Where a person is famous and has a marketable personality, any use of that personality for commercial purposes without their permission can constitute a misappropriation of personality. In this regard, the definition of “personality” can be very broad and include indicators such as one’s name, image, voice, jersey number, silhouette, etc. The key characteristics evocative of a famous personality often depend on the source of that person’s fame.
As established in the 1973 decision of Krouse v Chrysler Canada Ltd et al, the tort of misappropriation of personality has three elements. In order to establish the tort, a plaintiff must establish that:
- their personality was exploited by the defendant for a commercial purpose;
- the defendant’s use of the plaintiff’s personality is clear enough to identify the plaintiff; and
- the defendant’s use of the plaintiff’s personality suggests or implies an endorsement by the plaintiff.
Importantly, unlike the tort of breach of privacy, the tort of misappropriation of personality does not require the plaintiff to establish that the defendant’s actions were intentional or reckless. Indeed, in several cases, defendants were found liable even though they did not intend to evoke the plaintiffs’ personalities. Such cases suggest that the use of AI could lead to the same results.
In Athans v Canadian Adventure Camps Ltd et al, a summer camp was found liable for misappropriating an athlete’s personality on the cover of its promotional brochure. The summer camp offered water skiing and initially sought to partner with Mr. Athans, an internationally renowned professional water skier. However, Mr. Athans was unavailable. For that reason, the camp did not intentionally feature Mr. Athans’ name or likeness to promote the camp. Indeed, it even sent Mr. Athans a courtesy copy of its final brochure.
However, the marketing company that created the brochure was unfamiliar with water skiing and used photographs provided by the summer camp (including a photograph of Mr. Athans) to create a handmade line drawing of a water skier in action. When completed, the drawing strongly evoked Mr. Athans’s “trademark” pose of executing a turn with perfect form. The judge found the drawing to be strikingly similar to the principal photograph used by Mr. Athans to promote himself, including on his letterhead, his business card, in magazines and other publications. The image was arguably one of the most iconic and recognizable images in the waterskiing community. The camp apologized and stopped using the drawing.
In its decision, the Court found that since the brochure did not mention Mr. Athans at all, no reasonable person would wrongly believe that he was associated with the camp and, without any evidence of confusion, the camp was not liable for passing off. However, the Court concluded that the camp nevertheless misappropriated Mr. Athans’s personality for commercial gain and ordered damages. While Mr. Athans was famous only among a niche audience of water skiers, the Court noted that the camp was targeted to water skiers, who may recognize him.
This scenario highlights an obvious AI risk. Most AI models are trained on real-world data sets, including texts, photographs and videos. The more prominent a person is in the real world, the greater the weight that an AI model may attribute to them. If a user enters an AI prompt like “image of a water skier in action”, the AI could readily produce a result that is highly similar to an iconic photo of a famous water skier, for example. The resulting image may thereby evoke a famous personality without any knowledge or intent on the part of the person generating the content. Similarly, an AI prompt like “sing our slogan with the voice of an angel” could produce a voice imitating a famous musician described as having “the voice of an angel”. In that regard, AI is now regularly able to create voices and videos indistinguishable from those of a real person, including the following recent examples:
- Why fake AI calls impersonating US officials are ‘the new normal’
- OpenAI stops ‘disrespectful’ Martin Luther King Jr deepfakes
- AI can steal your voice, and there’s not much you can do about it
Conclusion
Using a real person’s image, voice or likeness in commercial products without permission creates legal risk of both breach of privacy and misappropriation of personality. Since AI models are able to mimic a person’s voice or likeness or to produce elements of real-world video or photographs unintentionally, using such models to create marketing material brings with it a heightened risk of liability. Brand owners should be careful to use AI models in a manner that does not unintentionally incorporate real people into their commercial uses.
The preceding is intended as a timely update on Canadian intellectual property and technology law. The content is informational only and does not constitute legal or professional advice. To obtain such advice, please communicate with our offices directly.