One week after a critical vulnerability in PHP knocked a swathe of websites offline, the German Greens have put a motion before the Bundestag that would make binding security standards mandatory for powerful AI systems — and make companies and their executives liable for the damage those systems cause.
The motion, reported on 27 October, calls for independent audits and a mandatory, documented post-mortem after serious incidents. "The safety of AI systems can no longer be left solely to the companies that develop them," said Rebecca Lenhard, the party's digital policy spokesperson, according to the report. Those who cause damage must be liable, she argued — and that should include the leadership of the large AI companies.
The timing is deliberate. The PHP flaw forced numerous sites to be taken offline as a precaution, and Lenhard pointed to it as evidence of how exposed digital infrastructure is. Her further argument is the one that turns a patching story into an AI story: attackers can use AI to find and exploit such holes ever faster, which the Greens read as a reason for tighter security rules for the most capable systems.
The motion also demands that the Federal Office for Information Security (BSI) and the AI Security Institute (AISI) be expanded in staff, funding and technical capacity, and that Germany push for common international AI safety standards, including at the United Nations. Greens MP Konstantin von Notz pointed to the incidents of recent weeks and said the government must show that announcements are followed by action.
The legal backdrop makes the demand more concrete than it sounds. The EU's revised Product Liability Directive explicitly covers software and AI systems as products, and Germany must transpose it into national law by 9 December 2026 — a deadline that turns "liability for AI" from a slogan into a drafting exercise with a date attached. The separate EU AI Liability Directive, which would have addressed exactly the evidentiary problems that make AI damage hard to litigate, was withdrawn in February 2025, leaving questions of proof and jurisdiction to national law.
That gap is the weak point of any liability regime, and the Greens' proposal does not yet say how it would be closed. Between manufacturer, operator and user of an AI system, responsibility can be apportioned in many ways; a chatbot that gives harmful advice, an agent that deletes a database, and a model that helps someone find a vulnerability are three different causal chains. Personal liability for executives is a further step, and one that would have to survive both German corporate law and the practical difficulty of proving that a specific management decision caused a specific harm.
There is institutional pressure from the other side too. BSI president Claudia Plattner recently noted that companies are deploying AI at speed while many still have no fixed procedures for AI security incidents — an operational gap that more staff and faster warning processes at the agency are meant to narrow.
The international context is moving in the same direction. Australia's cyber authorities warned on 27 October that AI is being used by state and criminal actors to exploit vulnerabilities faster and at greater scale.
