eco Survey: Companies in Germany Are Deploying AI Agents Under Controlled Conditions – Blanket Calls to Slow AI Development Miss the Mark

• 54% of the companies surveyed are already using AI agents or currently testing them
• 74% would allow AI agents to perform no more than supporting tasks or work independently under human supervision
• Only 5% would permit fully autonomous decision-making

Berlin, 14 October 2026 - Warnings about a loss of control and calls for stronger safety mechanisms are currently shaping the debate surrounding powerful AI systems. The focus is increasingly on how to harness the opportunities offered by ever more capable AI without relinquishing control and responsibility.

A recent survey conducted by the market research institute Civey among 500 companies on behalf of eco - Association of the Internet Industry shows that companies are already addressing this challenge in practice by setting clear boundaries and implementing safety mechanisms.

More than half of the companies surveyed are already using autonomous AI agents or testing them. At the same time, companies are placing clear limits on what these agents are permitted to do: 74% would allow AI agents to perform no more than supporting tasks or work independently under human supervision. Only 5% would consider allowing fully autonomous decision-making.

"The dystopian scenarios surrounding AI agents that are currently so widespread are out of touch with the reality of German companies. Companies in Germany want to use AI, but they do not let it make decisions blindly. This is what responsible use of the technology of the future looks like," says Oliver Süme, Chair of the Board of eco Association.

More than half of companies are using or testing AI agents

The use of AI agents is already a reality in many companies. Of those surveyed, 25% use AI agents in individual departments, while a further 11% have already deployed them company-wide. Another 18% are currently testing AI agents or conducting pilot projects.

This means that 54% are already using or testing AI agents. A further 7% have concrete plans to introduce them, while 33% currently have no plans to do so.

At the same time, companies clearly distinguish between using AI and handing over decision-making authority. Some 40% would limit AI agents to supporting tasks, while a further 34% would allow them to work independently, provided human oversight remains in place.

Another 13% would permit autonomous decisions within specific subprocesses. Only 5% would be prepared to allow AI agents to make decisions entirely autonomously.

"The figures show that companies in Germany want to harness the potential of AI agents while acting responsibly. They are gaining practical experience, establishing reliable control mechanisms and gradually making AI an integral part of their business processes. It is precisely this practical approach that brings innovation and trust together," says Süme.

Companies are implementing safety mechanisms

The companies surveyed are also focusing on specific safeguards to maintain control over AI agents.

Some 47% would introduce an automated emergency shutdown mechanism when deploying AI agents, while 45% cite strict access rights and role-based permissions.

Regular security audits would be considered by 37%, continuous monitoring of AI outputs by 32% and transparent activity logs by 31%. Multiple responses were permitted.

From eco Association's perspective, the findings show that companies are not ignoring the risks associated with AI agents but are considering safety mechanisms from the outset.

The objective is not to artificially limit the technology's capabilities, but to ensure that its use remains controllable.

Responsible deployment instead of a blanket slowdown

In eco Association's view, the findings should also inform the current debate about the pace and limits of AI development.

Caution when deploying specific applications should not be confused with fundamental scepticism towards the technology. Companies are introducing new systems gradually, testing applications and restricting their scope of action where they are unwilling to transfer responsibility entirely to machines.

"Safe AI is achieved primarily through responsible use, clear control mechanisms and practical experience. The more companies use AI effectively and learn from its application, the better we can harness opportunities and keep risks under control. We should accelerate this learning curve," says Süme.

Survey methodology

The market research institute Civey conducted an online survey of 500 companies on behalf of eco Association between 18 September and 3 October 2026.

The results are representative due to quota sampling and weighting, subject to statistical margins of error. For the overall results presented here, the margin of error ranges from 9.1 to 9.4 percentage points, depending on the question.

Further information on the methodology is available here.

The full survey results are available here. (German -language)