The rise of artificial intelligence has significantly impacted intellectual property offices, which act as barometers of technological enthusiasm. Registering a trademark is essential to secure market position, but a recent report warns of the risks of not adequately protecting AI projects: nearly two out of three trademark applications related to AI (63.66%) have deficiencies in their legal strategy. Many companies incorporate the term AI or its English acronym into their brands to leverage media appeal and search engine positioning. In 2025, there were 272 trademark applications in Spain that explicitly or implicitly reference artificial intelligence, according to data from the OEPM and EUIPO collected in the report “Spanish Trademarks Referencing Artificial Intelligence: Analysis of Applications 2025 and Gaps in Protection Strategy” by Pons IP. The analysis reveals that most holders protect their trademarks only in classes related to software and technology, neglecting protection in the specific sectors where they operate. This opens the door to competitors and increases the risk of application denial. The acronym AI predominates (44.1%) over IA (32.4%), indicating an international orientation of many Spanish projects, where English is the dominant language in the tech field, according to Manuel Campanero, director of the Pons IP office in Andalusia. 75% of applications were submitted to the OEPM and 25% to the EUIPO, reflecting the prevalence of start-ups and SMEs, which often prioritize national protection for cost reasons or lack of awareness of the advantages of community trademarks. The OEPM highlights that trademark applications reflect economic reality and that since 2020, there has been a notable increase in registrations related to AI. For example, in class 7 of Nice (machines and tools), 10 of the 11 AI-related applications were submitted from 2020 onwards. In class 9 (software and scientific devices), 1002 of 1229 applications were registered since 2020, and in class 42 (technological services), 867 of 938. A common strategy is to phonetically add the acronym IA at the end of the brand (16.5%), as in hostelerIA or TuristIA, facilitating its association with the sector. However, this can reduce distinctiveness, as occurred with the denial of the OpenAI trademark at the EUIPO in 2024, although the OEPM accepted its application, which is still pending. To overcome distinctiveness issues, it is recommended to opt for figurative versions of the brand, incorporating graphic elements that provide uniqueness, as OpenAI did at the EUIPO. The Pons IP report indicates that most applications only seek protection in classes 9 and 42, leaving other sectors unprotected. 63.6% of applications have some gap, 27.6% have not requested technological protection, and 36% have not managed sectoral protection. Only 36.4% have secured both aspects. Sectoral classes, such as 36 (financial services), 44 (health), or 45 (legal), have very low presence. Campanero warns that AI is a tool applied to specific sectors, so limiting protection to software and technological services is insufficient. The study shows that only 8 trademarks (2.9%) include class 45, despite the legal sector being one of the most active in AI solutions in Spain in 2025. Additionally, protection must go beyond the choice of Nice classes, considering the complex nature of these projects, which combine technology, data, third-party models, and user interfaces, according to Pablo López, director of brands & brand intelligence at Pons IP. López suggests protecting the source code as a trade secret, as the protection of the trained AI model is more complex and European jurisprudence is still unclear on this matter. Therefore, he recommends combining industrial property and trade secret. Many AI projects do not develop their own models but adjust base models from third parties like GPT or Gemini. Non-compliance with the terms of use of these models can lead to contractual claims or actions for unfair competition, so reviewing the terms of use is essential in the legal strategy of any AI product.