Managing AI Risk in Your Enterprise
The key sources of risk when using AI—and the measures that limit them
For management and heads of IT in mid-sized companies that use AI applications in their operations—but do not develop them.
Real-World AI Incidents
Anyone who does not know which tools are used and what data is entered can neither assess nor manage the operational risks. Real-world incidents show that these risks are not just theory:
- Air Canada: In 2024, the airline was held responsible for incorrect information given by a customer service chatbot; the competent Canadian tribunal (Civil Resolution Tribunal, British Columbia) rejected the argument that the chatbot was a separate entity.
- Cologne Local Court: In 2025, a lawyer filed a brief containing invented, apparently AI-generated rulings and citations. The court found this and instructed him to refrain from doing so in the future (Ref. 312 F 130/25).
- IBM study: One in five of the companies examined by IBM reported a data breach caused by unapproved AI. Companies with a lot of unapproved AI paid on average USD 670,000 more per data breach than companies with little or none (IBM 2025).
The vulnerability rarely lies in the AI model itself, but in organizational gaps—such as access rights that are too broad, unchecked work results and missing guardrails. The legal risks already exist today—through the GDPR, the German Trade Secrets Act, the German General Equal Treatment Act and, for companies covered by NIS-2, the German BSI Act (Sections 30 and 38 BSIG).
Inventory AI Usage. Assess Risks. Determine Mitigations.
- Inventory AI usage: The template records for each use of AI which tool is used by whom with what data—including those that have not been approved. The completed table is at the same time your AI inventory.
- Assess risks: The whitepaper assigns the key risk sources of AI use to four areas—Input, Access & Execution, Output and Ecosystem. The first three map the lifecycle of an AI use case—from the information entered, through the AI's access and actions, to its results. The Ecosystem forms the environment of technical, organizational and external dependencies. Risk sources covered include data leakage, indirect prompt injection, flawed or manipulated data, oversharing, hallucinations, automation bias, shadow AI, and manipulated third-party AI components.
- Determine mitigations: For each risk source, the whitepaper names recommended measures. Which measure fits depends on the individual use of AI.
Questions the Whitepaper Answers
- Which AI tools are actually used in your company—including without approval?
- What risks arise when data is entered, when the AI accesses systems and from its results?
- Which measures limit these risks?
- Which legal obligations already apply—for example under the GDPR, German trade secret and equal treatment law and the BSIG—and which are coming?
- How do you carry out your first AI risk inventory yourself?
Template: Your First AI Risk Inventory
The Excel template contains drop-down lists and the check questions for each risk source. Every use of AI in the company is recorded—including those that have not been approved. How to proceed:
- Collect: IT surveys all business units on which AI tools they use—explicitly including private and free ones. In addition, IT checks four sources: firewall or proxy logs, the Microsoft 365 admin center (or the administration of comparable platforms), browser extensions on company devices, and expense reports and company credit cards.
- Record: IT logs each entry together with the business unit concerned. The check questions show which risk sources apply to this use of AI.
- Assess and decide: IT presents the entries with the highest risk to management and recommends measures. Management decides which measures are implemented and who implements them.
Download the Whitepaper and Template
Free of charge and without registration.
Your Next Step: Shadow AI Check With AI Inventory
Together with your head of IT, I record which AI tools are actually used in your company—including those nobody has approved. The result is an AI inventory: which tools, in which departments, with which data. The AI inventory feeds into a later AI Compliance Dossier or AI management system.
More about this service: Shadow AI Check with AI Inventory
About the Author
Derek Carson—AI governance and information security, Künzell near Fulda, Germany. Passed the Lead Implementer exams for ISO/IEC 42001 and ISO/IEC 27001 · ISC2 Certified in Cybersecurity. Native English speaker, working in English and German. More about me
Another whitepaper: Implementing NIS-2 on the Basis of ISO 9001