More Than 100 Companies Warn of Cyber Risks from AI Agents

Structured Proposal from the Business Community

According to the publications, the signatories include several major technology corporations and security firms. Among those named are OpenAI, Anthropic, and SAP. The research and security community is also involved. The breadth of the alliance is intended to show that the risks no longer affect individual sectors alone, but the entire digital economy.

The concern is directed primarily at AI agents – software that plans and executes tasks autonomously. While classical automation tools follow rigid rules, such agents can make decisions, combine tools, and adapt to new situations. It is precisely these qualities that, in the authors' view, make them attractive for attackers – for example, in finding vulnerabilities or in the automated exploitation of known security gaps.

Jürgen Schmidt, a security expert at a participating company, told journalists that today's defense mechanisms are designed for static threats. "Wir sehen Angriffe, die in Minuten ablaufen und nicht in Tagen," sagte er. That changes the rules of the game and requires a different speed in detection and response.

Why AI Agents Change the Threat

The authors of the letter point out that the underlying AI models are becoming increasingly powerful. This also increases the potential to use them against one's own infrastructure. An early examination of possible vulnerabilities is therefore not an option, but a necessity. At the same time, they emphasize that the same technologies could also be used for defense.

Specifically, the signatories are calling for, among other things, binding security standards for AI agents, mandatory testing before the deployment of such systems, and better information sharing between research, authorities, and industry. In addition, the training of cybersecurity professionals should be significantly expanded to reduce the existing shortage of skilled workers.

The letter also formulates time pressure: the current window to develop binding rules and implement technical protective measures is limited. If it is missed, there is a risk of a phase in which attacks scale before defense systems can catch up.

Concrete Demands to Policymakers and Business

Observers see the initiative as one of the broadest industry responses to date to AI-specific cyber risks. Compared with earlier statements, which usually came from individual companies or research groups, the appeal brings together actors from business, research, and security agencies.

The signatories announced that they would bring their proposals into political processes in the coming weeks. These include discussions with regulatory authorities as well as contributions to ongoing standardization procedures. However, it remains open how binding such standards will ultimately be.

In a second step, the alliance plans to launch concrete pilot projects to test how AI agents can be monitored and contained in controlled environments. Findings from these are to be incorporated later into public guidelines.

Pilot Projects as the Next Step

Overall, the authors paint a picture of a technology that magnifies both opportunities and risks. Their appeal is therefore directed not only at policymakers but also explicitly at companies of every size to review their own AI and security strategies.