Amid the European Commission’s public consultation seeking input on a possible legislative initiative on copyright and AI (amongst other topics – see our previous post), several Member States are seeking to forge ahead with their own national ‘fixes’.
These national initiatives are potentially important bellwethers for rightholders and users alike, offering an early indication of the approaches that may ultimately shape the debate, and potentially even the legislation, at EU level.
A new right for digital replicas?
After Denmark and the Netherlands, it is now Belgium’s turn to look to copyright and neighbouring rights law as a means of protecting a person’s image and voice against ‘deep fakes’. For the Belgian MPs behind the proposal, the AI Act’s rules on deep fakes do not go far enough. Those rules only require labelling and do not restrict the dissemination of unauthorised deep fakes. The proposed definition of deep fake is, however, closely modelled on the AI Act: any AI-generated or manipulated image, video or audio content that closely resembles a living or deceased person and could falsely appear to be authentic or truthful would fall into scope.
The proposed right would operate like a neighbouring right rather than as a free-standing personality right. It would grant natural persons – including for fifty years after their death – a right to object to the creation, reproduction or communication to the public of deep fakes incorporating their image or voice. Building on that protection, the bill would create a licensable right over a person’s likeness for AI-related uses, subject to Belgian copyright contract rules.
The proposal also deals directly with performers’ rights. According to the Belgian MPs, existing performers’ rights only cover reproductions of fixed performances and do not provide a directly actionable right against ‘synthetic imitations’. The proposal would therefore extend performers’ neighbouring rights to AI-generated imitations of performances.
Deep fakes used for ‘parody, satire or social criticism’ would be excluded from the scope of the new rights, provided that, in light of all the circumstances of the case, a fair balance is struck between the right to respect for the person’s image and voice and freedom of expression.
Like its Danish and Dutch counterparts, the Belgian proposal seeks to repurpose a framework originally designed to protect performers recorded performances and to ensure that they are remunerated for those creative contributions, turning it instead towards the protection of the attributes of natural persons, a function already served by other legal frameworks. It is not clear whether neighbouring rights can accommodate such a fundamental shift without undermining their core functions: protecting performers’ creative contributions and enabling those contributions to be incorporated into collective productions such as audiovisual works and phonograms. A new right would disrupt established licensing practices by adding a new layer of rights clearance.
Indeed, this highly unusual melding of personality rights-concerns and copyright law may be well-intentioned, but it could create significant complications in the creative industries, in particular for the AV sector.
Presumption of use of protected works by AI providers
A second group of Belgian MPs has tabled a separate AI-related copyright proposal. This proposal, which would introduce a presumption that copyright protected content has been used by an AI provider in certain circumstances, comes hot on the heels of a similar bill adopted by the French Senate in April 2026. However, that bill has since stalled in the Assemblée Nationale (the lower chamber).
The Belgian proposal would introduce a new IP enforcement rule, shifting the burden of proof to AI providers in copyright infringement proceedings brought against them. The presumption would arise where rightholders provide plausible evidence of such use. Such evidence could take any form, including outputs reproducing the works, prompts and training data summaries and material disclosed in judicial proceedings abroad (presumably in places like the US and UK which have far more extensive discovery practices).
Crucially, the presumption relates only to the fact of use, not to its lawfulness. In other words, it does not determine whether that use infringes copyright or falls within an exception. AI providers could rebut the presumption by demonstrating either that the claimant’s works were not used or that any use fell within an applicable copyright exception.