Executive summary
Advancements in generative AI have simplified the process of generating new and derivative creative works by ingesting and reconstituting "personhood signals" – voice timbre, phrasing, visual likeness, performance style – that make individual artists recognizable. Traditional intellectual property rights were not built for this problem: copyright struggles to prevent model training, and the applicability of fair dealing (in Canada) / fair use (in the US) remains unclear.
In response, US artists (most visibly singers, but also other actors and a growing roster of athletes and influencers) are stretching the non‑traditional trademark toolkit, including sound marks for short voice clips, motion marks, and visual marks describing on‑stage likeness, to convert recognizable identity features into registrable brand assets.
Canadian law accommodates the same categories of non‑traditional marks, and a review of the Canadian register shows such marks (e.g., sound marks, motion marks, 3D guise, position marks) are already in active use.
For in‑house counsel advising artists and other creators, or talent‑adjacent businesses with Canadian exposure, non‑traditional trademarks are potentially a comparatively inexpensive, defensive complement to copyright and personality rights. They can be useful particularly for downstream enforcement against AI‑generated outputs that exploit identity in commerce, even if they cannot meaningfully police the training stage itself.
Uncertainty remains regarding the enforceability of these non-traditional trademarks due to a lack of jurisprudence and guidance.
Limits of traditional IP against ML training and generative outputs
The first challenge is machine‑learning training is fundamentally an internal, non‑consumer‑facing act of copying. Trademark liability typically turns on a defendant's use of a mark in commerce in a way likely to confuse consumers as to source, sponsorship, or affiliation, and that framing does not naturally reach the ingestion of recordings, lyrics, or images into a training corpus. Copyright is the more relevant IP right at the training stage.
The second challenge arises at the output stage. Generative AI systems do not merely copy; they recombine learned patterns to generate outputs that "may not be copies in the traditional sense, but can function as substitutes," such as synthetic vocals that sound like a particular singer, images that read as a particular performer, or text that mimics a particular author's cadence (e.g., the iconic “beatnik” cadence).
Copyright struggles where the output is not a literal reproduction of any one work, and trademark struggles where no registered mark has been "used" in the classical sense. The result is a doctrinal gap precisely at the place where identity itself is being commercially exploited.
US response: Stretching non‑traditional marks to cover identity
US practitioners have begun closing this gap by leaning on non‑traditional trademarks. In April 2026, a well-known singer’s management entity filed three applications at the United States Patent and Trademark Office (USPTO): two sound marks for short spoken phrases and a visual mark depicting the singer on stage holding a pink guitar and wearing a multicolored iridescent bodysuit and silver boots.
The strategy is experimental: trademark law has long protected source‑identifying sounds, but whether registration of a celebrity's spoken phrase can be enforced against AI‑generated vocal imitations that evoke the artist's identity without copying the exact clip have not been fully tested in the courts.
What these filings are really trying to protect is the set of "personhood signals" that AI systems readily ingest and audiences most readily recognize: voice, phrasing, timbre, performance style, and visual persona.
Trademarks may not stop training, but at the output and commercialization stage, synthetic endorsements, fake promotions, AI‑generated merchandise, registered voice clips, image marks, and stylized name marks give artists protectable brand assets that exist independently of copyright in any underlying work.
USPTO Director John Squires noted in his International Trademark Association 2026 keynote that the agency is seeing a "subtle but unmistakable shift" toward filings tied to "Voice. Persona. Motion. Authenticity," with actors / actresses filing for sound and motion marks and athletes filing trademarks before turning professional, all driven by AI‑accelerated risks including deepfakes, voice cloning, and synthetic endorsements.
Director Squires described protection of name, image, and likeness of individuals as the "frontier issue" for trademark eligibility, sitting at the intersection of state right‑of‑publicity laws and federal Lanham Act protection.
Mapping the strategy onto Canada
Canadian law is well structured to enable artists to adopt the US approach.
Since the June 17, 2019, amendments, the definition of "trademark" expressly includes "a sign or combination of signs," and the statutory list of signs covers sounds, three‑dimensional shapes, holograms, moving images, modes of packaging, scents, tastes, textures, colours, and the positioning of a sign.
Sound marks, 3D shapes, colour marks, and several other non‑traditional categories require evidence of distinctiveness at the time of filing, which is a meaningful, but not insurmountable, hurdle for well‑known individuals.
Canada's name‑mark rules add a wrinkle US counsel often underestimate. Under paragraph 12(1)(a) of the Trademarks Act, a word that is "primarily merely" the name or surname of a living individual (or one deceased within 30 years) is not registrable absent acquired distinctiveness as of the filing date. Section 9(1) similarly restricts the adoption of marks consisting of, or closely resembling, the portrait or signature of such an individual without written consent.
Reviewing the Canadian Intellectual Property Office’s (CIPO) trademarks database for non‑traditional marks shows the infrastructure is already well exercised, though, at this time, almost entirely by corporate brand owners. There are specific registered sound marks for different start-up chimes, movie production studio opening logo sounds, jingles, which are all source identifiers built around a short, recognizable acoustic signal, which is conceptually the same or a similar construct singers are trying to use for their brand distinctiveness.
Motion marks such as the logo motion sequences, among others, can also be adapted to artist-based motion mark filings, noted USPTO Director Squires.
The closest analogue on the Canadian register to a celebrity‑identity 3D mark is a three‑dimensional/design registration for a distinctive stage guitar, which demonstrates that a recognizable performance prop or stage element can itself be carried as a 3D/design mark in Canada.
Conclusion
This is an evolving area, and material uncertainty remains. CIPO has not yet had to decide how aggressively to police acquired‑distinctiveness evidence for voice‑clip sound marks tied to a single performer, and Canadian courts have not tested whether such a registration can be enforced against AI outputs that evoke, but do not literally reproduce, a registered clip.
It is unclear whether trademark law will be able to reach the training act itself, and personality‑rights doctrines (the Canadian analogue to right of publicity) should be considered a better fit for many identity‑related torts.
Even so, non‑traditional trademark filings in Canada are comparatively inexpensive, federal in scope, and durable, and they create a portfolio of registered identity assets that can be deployed at the output and commercialization stage against synthetic endorsements, AI‑generated merchandise, and unauthorized branded uses of a client's persona.
For counsel advising entertainers, athletes, influencers, or the brands that license them, the practical takeaway is to start considering sound, motion, 3D, position, and stylized signature filings as a core layer of protection of image, and likeness of individuals in Canada, alongside copyright registration, personality rights, and contractual controls.