Earlier in May, the European Commission announced that the AI Act could undergo targeted amendments. This is part of the EU's plans for AI innovation through the AI Continent Action Plan, which emphasizes regulatory simplification. Previously, the plan focused on establishing an AI Act Service Desk to assist with compliance rather than amending the act itself.
Reports suggest that the Commission might delay the application of the AI Act. Although it came into force in August 2024, most provisions, including those on high-risk and general-purpose AI systems, are not yet applicable.
The issue of intellectual property rights in generative AI has gained attention due to two institutional reports and pressure from Member States regarding copyright concerns in general-purpose AI (GPAI) models. The European Union Intellectual Property Office (EUIPO) highlights that generative AI training increasingly uses copyrighted content and stresses the importance of an opt-out mechanism under article 4 of the DSM Directive. The report calls for better technical solutions and standardization practices to address copyright issues but does not clarify what constitutes a valid "machine-readable" opt-out.
The European Parliament Research Service (EPRS) criticizes current EU copyright rules as inadequate for GPAI training due to data collection scale and opacity. It notes ongoing legal uncertainty around text and data mining exceptions and suggests that the 2026 review of the Copyright Directive could address these gaps.
Portugal, Spain, and Italy have urged stronger copyright protections within GPAI models ahead of a Council meeting in Brussels. They advocate for transparency measures like unique identifiers to track copyrighted content use in AI models.
On May 20th, a public hearing titled “Empowering creators in the AI Age” was held by the EU Parliament to discuss challenges posed by AI to European culture. Participants included representatives from DG Connect, DG EA, and CISAC President Björn Ulvaeus.
The European Commission published an analysis of stakeholder responses regarding definitions and guidelines for prohibited AI practices. The report notes industry-dominated responses may not fully represent those most affected by AI systems. Stakeholders seek clarity on whether traditional software should be excluded from the act's scope but warn against potential loopholes exploited by providers reclassifying their systems as traditional software.
In recent developments related to EU laws on AI, ChatGPT might be designated as a systemic search engine under the Digital Services Act (DSA). If designated as such upon reaching 45 million active monthly users, it would face higher obligations reserved for very large online platforms or search engines.
Other news includes a consultation on Data Union strategy aiming at building interoperable datasets necessary for AI development. The European Commission also seeks third-party contractors for assessing risks posed by GPAI models through a forthcoming tender divided into six lots focusing on different systemic risks.
Furthermore, a German court rejected an injunction against Meta concerning user data usage for training its AI models. Meanwhile, noyb sent Meta a cease-and-desist letter over its plans to use personal data from Facebook and Instagram users without proper legal basis under EU law.
Lastly, Italy fined Luka Inc., operator of chatbot Replika €5 million for GDPR violations related to inadequate privacy policies.
CDT Europe recommends further reading on topics such as GDPR reforms potentially unraveling protections or how Big Tech influences advanced AI regulations.