The European Union is considering measures to slow the rapid development of the world’s most powerful artificial intelligence models as concerns grow over cybersecurity, misuse and the possibility of humans losing control over increasingly autonomous systems.
European Commission President Ursula von der Leyen said she plans to invite leading developers of frontier AI models to discuss the risks associated with the technology and explore safeguards that could be adopted before more capable systems are released.
The initiative would focus on advanced general-purpose models capable of performing complex tasks across multiple fields. These systems can generate software, analyse scientific data, conduct research and interact with other digital tools with decreasing levels of human supervision.
European officials are particularly concerned that future models could be exploited by hostile states, organised criminal groups or individual attackers to conduct sophisticated cyber operations. Powerful AI systems might help identify software vulnerabilities, automate hacking campaigns or develop malicious code at a speed that overwhelms existing defences.
Another concern involves the growing autonomy of advanced models. Researchers have warned that highly capable systems could behave in unexpected ways, circumvent restrictions or pursue assigned objectives using methods their developers did not anticipate.
The proposed talks would bring together major AI laboratories, regulators, cybersecurity specialists and independent researchers. Discussions are expected to cover model testing, security evaluations, controlled access to sensitive capabilities and procedures for delaying deployment when a system fails essential safety assessments.
Any slowdown would not necessarily amount to a complete suspension of AI research. Instead, Brussels could seek agreements requiring companies to conduct more extensive safety testing before training or releasing significantly more powerful models.
Such measures could include independent evaluations, stronger protection for model data and computing infrastructure, mandatory reporting of serious incidents and greater transparency about the capabilities and limitations of frontier systems.
The European Union has already positioned itself as a leading global regulator of artificial intelligence through its AI Act, which establishes different obligations according to the level of risk posed by an AI application. However, the rapid development of frontier models has raised questions about whether existing rules can keep pace with technological progress.
Technology companies may resist measures that significantly delay development, arguing that excessive restrictions could weaken Europe’s competitiveness against the United States and China. Industry representatives are also likely to warn that unclear regulations could discourage investment and push innovative companies outside the European market.
Supporters of stronger safeguards maintain that competition cannot be allowed to override public safety. They argue that developers should demonstrate that their systems are secure and controllable before releasing models with capabilities that could affect critical infrastructure, national security or millions of users.
The Commission has not yet presented a formal proposal or timetable. The planned consultations are expected to help determine whether voluntary commitments are sufficient or whether the EU should introduce additional binding rules for developers of the most advanced AI systems.
The debate highlights Europe’s attempt to balance technological innovation with security and democratic oversight as artificial intelligence becomes more powerful, autonomous and deeply integrated into modern society.
