The Higher the Stakes, the Tighter the Reins: Only 2 Per Cent of Companies Allow AI Agents to Handle Money Independently

AI agents can carry out tasks independently, but companies retain human oversight precisely where money and critical data are concerned. These are the findings of a survey of 500 German corporate decision-makers commissioned by the IT service provider adesso. The agents’ freedom decreases as the importance of a task increases: according to 38 per cent of respondents, agents are allowed to carry out research without human approval. For customer enquiries, the figure is 19 per cent. When it comes to orders and financial transactions, the figures drop to 8 per cent and 2 per cent respectively.

Research Yes, Finances No: Trust Based on Risk

Respondents also keep a tight rein on data in business-critical systems: only 4 per cent allow AI agents to modify this data without approval. A look ahead confirms the pattern: by 2028, 74 per cent of respondents expect agents to be permitted to research and process information without approval; for customer enquiries, the figure is 55 per cent. When it comes to data in business-critical systems and financial transactions, however, around 60 per cent in each case expect that agents will still require human approval even by 2028. Despite this somewhat cautious assessment, the results reveal a clear trend towards greater trust in autonomous AI agents; they are increasingly becoming cornerstones of business processes – even in sensitive areas.

Control and Accountability Lag Behind Usage

To ensure that agents only carry out authorised actions without approval, they must be subject to oversight. This is where gaps become apparent: only 34 per cent of respondents state that their company can stop an agent immediately, and only 29 per cent can reverse erroneous actions. Full logging of agent actions is in place at 35 per cent of companies, whilst 27 per cent are unable to assess what control options their company has. Responsibility is also rarely defined in a clear and binding manner: only 19 per cent have fully defined who is responsible for an agent’s operation, monitoring, further development and potential errors. The results can be interpreted as follows: decision-makers distribute their trust in AI agents according to risk.

Competition: Decision-Makers Rate Germany’s Position in AI as Merely ‘Adequate’

Respondents also assessed Germany as a location. Here, their verdict was critical. On average, they rated Germany’s AI competitiveness as ‘adequate’ and that of the EU as ‘satisfactory minus.’ The US and China, by contrast, are in the ‘good’ category, with China just ahead of the US. Only 8 per cent rated Germany as ‘very good’ or ‘good’, compared with 67 per cent for the US and 70 per cent for China.

Company size also influences the importance attached to AI: 27 per cent of respondents from companies with 500 or more employees consider it to be of great importance for competitiveness. In small and medium-sized enterprises (SMEs) with up to 499 employees, however, the figure is just 10 per cent.

The website www.adesso.de/moveAI shows how adesso supports companies in integrating AI and agents into live operations with clear rules and human accountability. The key findings of the study are also available there.

Methodology: For the study, the market research institute Heute und Morgen conducted an online survey of 500 German corporate decision-makers in early October 2026, including business owners, board members, managing directors and senior executives from all sectors. The survey does not claim to be statistically representative. The questionnaire, results and further charts are available on request.