The widespread adoption of generative artificial intelligence (Gen AI) in Croatia’s tech, media, and professional services sectors is raising a simple but crucial question: who owns the outputs created by these tools, and what can users legally do with them?
While the impressive capabilities of tools like ChatGPT, DALL·E, Suno, and others have been extensively discussed, there is a significant lack of public awareness about the contractual terms that govern their use. These terms, embedded in lengthy service agreements, carry real legal and commercial implications. In Croatia, where businesses and creatives alike are increasingly integrating AI into their workflows, understanding these terms is essential.
Most Gen AI providers claim to grant users ownership of outputs generated in response to their prompts. On paper, this sounds like a green light for unrestricted use and commercialization. In reality, however, such rights are frequently hedged with legal limitations and carve-outs that make full control more illusory than real.
Take Anthropic’s Claude, for example: although users are said to retain rights to their outputs, the company reserves the right to reuse both inputs and outputs to improve its systems and develop new products. It also bars users from leveraging the content to train competing AI systems or to support rival services. These restrictions limit how Croatian developers and start-ups can lawfully repurpose the content they generate, especially when considering local efforts to build competing AI platforms.
OpenAI, the company behind ChatGPT and DALL·E, applies even more detailed restrictions. Users must clearly disclose that any published content was AI-generated and are expected to follow a set of usage rules covering everything from prompt transparency to user attribution. Notably, OpenAI retains rights to use input and output data for system improvement and compliance purposes, although it allows users to opt out of data training. For Croatian businesses concerned with confidentiality or intellectual property protection – such as law firms, marketing agencies, and software developers – this creates uncertainty about how safe it is to feed sensitive content into these tools.
Meta’s Llama model offers a broad license to use, modify, distribute, and even create derivative works based on both inputs and outputs. However, at the same time, users must not infringe on the rights of others and must accept that their data may be reviewed or reused in research contexts. Requests to delete content can take up to 90 days. This could be particularly problematic in regulated Croatian sectors such as banking or healthcare, where timely data deletion is often a compliance requirement.
Midjourney, popular for generating digital art, gives users ownership over their creations, with some important exceptions. Companies earning more than USD 1 million annually must pay for a premium license to secure ownership rights, and any image that builds on someone else’s upscaled creation remains the property of the original user. This has implications for Croatian design studios and creatives using the platform for commercial work, especially those working with client commissions.
Suno offers a more rigid division between free and paid users. Only paying users may use outputs for commercial purposes, and even then, they may not use them in ways that compete with Suno’s services. Free users are limited to non-commercial, internal purposes with mandatory attribution. For Croatian musicians and audio producers experimenting with Gen AI for song writing or track creation, this can be a major limitation on monetization.
Synthesia’s terms reflect an organizational logic. Rights to the output rest with a company, not an individual employee using the service. This could confuse Croatian SMEs, where such internal ownership structures are not always formalized. Moreover, Synthesia explicitly states that its service is for commercial use only, meaning private users may not benefit from typical consumer protection laws under Croatian law.
Across all these examples, a clear pattern emerges: providers seem to shift both liability and responsibility to users. While users are nominally granted rights to AI outputs, they are also tasked with ensuring legality, IP compliance, and adherence to applicable laws. For users in Croatia, this means that using Gen AI tools without a thorough legal review of the terms of service can lead to unanticipated risks, particularly if the output is used for commercial or public-facing purposes.
Therefore, the prevailing myth that “you own what the AI gives you” does not hold up to contractual scrutiny. For Croatian businesses, creators, and legal advisors, navigating the legal grey zones of Gen AI output requires more than enthusiasm – it demands a cautious and attentive reading of the fine print to fully understand the implications.
By Dino Gliha, Partner, MGG Law
This article was originally published in Issue 12.7 of the CEE Legal Matters Magazine. If you would like to receive a hard copy of the magazine, you can subscribe here.
