AI security: experts warn of risks and call for better control mechanisms

The security of advanced AI systems is the focus of discussion after several experts and organizations issued warning signals. OpenAI CEO Sam Altman emphasizes that it is impossible to lock AI in security environments in the long term, as the problem of alignment – ensuring that AI goals match human values – has not yet been solved. At the same time, Jensen Huang, CEO of Nvidia, warns that AI developers cannot simply be regulated and instead calls for stronger technical security. He recommends OpenShell open source software as a seat belt for AI agents. These warnings are underpinned by an on-site inspection at OpenAI, where significant vulnerabilities in the AI surveillance system were discovered. The auditors found that safety measures were not sufficient to effectively minimise risks. At the same time, Microsoft is setting new standards with Copilot by running AI models in sandboxes, both locally on Windows systems and in the cloud. This concept of isolated execution is intended to increase the security and controllability of AI systems. The discussion about AI security is therefore not only technical, but also ethical and regulatory. It is about the question of how AI systems can be developed and used without endangering people and society. In this context, it is particularly important to find solutions that reconcile both innovation and security. The challenge is to design AI systems to respect human values while unleashing their full potential. This requires not only technical measures, but also a deep understanding of the ethical implications of AI developments. The security of AI is a complex topic that encompasses many facets – from technical implementation to social acceptance. It requires a variety of approaches to ensure sustainable and responsible AI development.