European experts emphasize that one of the main lessons from the EU process is that AI security must start from concrete risk scenarios, such as the use of technology for biological weapons, cyberattacks on government systems or threats to critical infrastructure.
Europe and the UK are developing their own approaches to testing and safety of artificial intelligence models, as the US administration prepares to impose new rules in this area. According to recent analysis, European countries have spent several years building a system of oversight for the risks posed by AI, while the US is still determining which models and uses should be included in a common framework.
President Donald Trump’s administration is expected to publish a voluntary framework for Artificial Intelligence, while also considering proposals from companies such as OpenAI and Anthropic to create uniform standards for the most advanced models, regardless of whether they are open or closed source.
European experts point out that one of the main lessons from the EU process is that AI security should start from concrete risk scenarios, such as the use of technology for biological weapons, cyberattacks on government systems or threats to critical infrastructure. Another important principle of the European approach is the use of standardized testing and independent verification, rather than companies relying solely on their own assessments of the security of models. Experts also argue that the EU has tried to base risk thresholds on scientific research and academic publications, while in the US standards often rely more on industry and self-regulation.
Another challenge facing the public sector is the exodus of technical experts to private technology companies, particularly in the US and UK. According to researchers, this could weaken governments’ capacities to control the rapid development of Artificial Intelligence.
In conclusion, Western countries are gradually moving towards a common understanding of the cyber risks of AI, but how the broader challenges of this technology will be managed still remains in the policy-making phase.