Berlin. The Green parliamentary group in the Bundestag is calling on the federal government to limit the security risks of powerful artificial intelligence (AI) more closely. This emerges from an application that is available exclusively to the Handelsblatt.
“AI systems are not only becoming more and more powerful, they are also taking on increasingly complex tasks independently,” it says. With advances in autonomous systems, the scope of possible risks is also growing. The federal government should therefore examine how autonomous AI systems can be safely switched off in an emergency.
The Green politicians also want to hold AI system providers more accountable in the future. Security incidents should be dealt with in a binding manner and clear liability rules should also apply. “The security of AI systems can no longer be left solely to the companies that develop them,” said digital politician Rebecca Lenhard, co-author of the application, to the Handelsblatt. The management levels of companies should not be spared. "Anyone who causes damage must be liable. This also applies to the management levels of the large AI companies."
The Green group accuses the large AI companies of shaping the public debate about AI security. Their managers set the tone, it says in the application. As a result, the debate revolves primarily around the performance of the systems, while gaps in security precautions and the responsibility of the companies themselves are lost sight of.
“Serious incidents already show that the leading AI companies are not adequately securing their systems,” said Green digital politician Lenhard. So far, companies have largely decided for themselves what risks they should take. “Given the potential consequences for society as a whole, this is no longer acceptable.”
To ensure that risks can be identified and limited at an early stage, the Green Party is calling for binding safety standards and independent audits. In the future, independent scientific findings should count as much as concrete incidents for political decisions, the paper says.
The request comes a week after numerous websites were taken offline as a precautionary measure due to a critical security vulnerability in the web programming language PHP. “The current PHP security vulnerabilities show how vulnerable our digital infrastructure is,” said Lenhard. With the help of AI, attackers can detect and exploit such vulnerabilities ever more quickly, while the time for countermeasures is shrinking, warns the digital politician.
“We must not act until a security gap has become a security crisis.” The population also needs to be better informed about AI-aggravated cyber risks and protection options.
The chairman of the parliamentary control committee, Marc Henrichmann (CDU), spoke of a “turning point” shortly after the security gap became known. Artificial intelligence quickly turns vulnerabilities that were previously considered hard to exploit into “real gateways”.
The Federal Office for Information Security (BSI) also sounded the alarm and issued an official cybersecurity warning. Hackers could launch massive attack campaigns and subsequently take control of websites or access sensitive customer data, it said.
The proposal from the Green parliamentary group therefore calls for the BSI to be expanded “in the short term and in line with needs”: in terms of personnel, finances and technology. This is the only way the authority can address the new security risks “that arise from more powerful and increasingly autonomous AI systems,” according to the application. The black-red coalition is already planning to significantly strengthen the BSI in the draft budget for 2027.
Green politician Konstantin von Notz welcomes the fact that the federal government has also recognized the importance of the topic and is making it a focus of the digital summit on November 2nd and 3rd. “The harsh incidents of the last few weeks and days have once again dramatically demonstrated the urgency of concrete political actions,” von Notz told the Handelsblatt. It is now up to the federal government to show that “noble announcements” are followed by actions.
The Green politicians also want to expand the German AI security institute AISI, which, as a national evaluation institution, assesses the capabilities and risks of powerful AI models and advises authorities and companies. They also advocate a close connection to the European AI Office. Digital Minister Karsten Wildberger (CDU) has already announced an expansion of the institute, but the federal government has not yet named any specific timelines or fixed funding for this.
The application also calls for the federal government to advocate internationally, especially at the United Nations, for common AI security standards and independent international supervision.
