Sagor warns AI must not cheapen actors’ work

Tuesday 1st September 2026 on 09:00 in Estonia

acting, artificial intelligence, copyright

Voice and digital likeness belong to individuals and must be protected as personal data, Reimo Sagor, chairman of the Estonian Actors’ Union and an actor at Vanemuine Theatre, said in an interview published by ERR and Sirp. He said artificial intelligence creates new opportunities but also systemic risks that could make acting more one-dimensional and less human.

“Agreements in the creative fields emphasise three principles: human work must be valued, payment for the use of artificial intelligence must be at least equivalent to what would have been paid for real work, and AI-generated content must be clearly labelled so audiences know when they are seeing or hearing machine-generated material,” Sagor said.

Sagor recalled a case from several years ago, when Ellen Kilgas recorded an audiobook about her father at his studio. She wanted to use sentences in Tõnu Kilgas’s voice, so she recorded his quotations, sent them to the Netherlands and had artificial intelligence create her late father’s voice from the samples.

Voice is part of a person’s personal data and belongs to the individual whose voice it is, Sagor said. Without knowing all the facts, he assumed that Ellen Kilgas was the heir to Tõnu Kilgas’s performers’ rights and had the right to decide how her father’s voice could be used, including whether to allow it to be reconstructed using AI.

However, legal and ethical boundaries must still be observed, he said. Inheriting rights does not automatically grant permission to create artificial derivatives of someone’s voice. Many people would probably like to know whether the voice of the late actress Ita Ever could be used to advertise a product, Sagor said, but this is not currently permitted. The voice of a deceased actor cannot be artificially restored and used for commercial purposes without a clear legal basis and the consent of the heirs.

Sagor said the creation of artificial voices remains a grey area in copyright law. He explained that AI systems break works down into tokens, the smallest units of text, and vectors, mathematical representations that describe an object’s properties as numbers. Although these are not literally works themselves, they contain metadata from the source material and can often be used to reconstruct the original.

“That is what distinguishes a machine from a person: a person imitates, while artificial intelligence reproduces,” Sagor said. He argued that the claim that a token is not protected by copyright is therefore unconvincing. If an embedding, a semantic vector describing an object’s content and its connections to other objects, makes it possible to reconstruct the original work, this amounts to hidden copying and raises serious copyright and ethical questions.

Sagor also referred to natural history filmmaker David Attenborough, whose distinctive voice is recognisable worldwide. He said he had seen random AI-generated clips on YouTube using the same vocal quality, immediately bringing Attenborough to mind.

“It is both funny and sad. He clearly did not record those texts, but his personal presence can be perceived through his voice,” Sagor said.

He said AI uses creators’ work and voices in ways that are effectively illegal, while legislation catches up only after damage has already been done. Technology is advancing quickly, but rules are changing slowly, making it difficult for performers and authors to defend their rights. Permission is not sought, voices and performances are used arbitrarily, and the system is structured so that violations are easy to commit and difficult to prove.

Source 
(via ERR)