If an AI provides the basis for business decisions, the people responsible are liable, not the machine. This makes the use of artificial intelligence risky for board members and management. At the same time, it can sometimes be unavoidable. The key question is therefore: how far does the business judgment rule apply when the information comes from a black box? And conversely, can the executive bodies be obliged to use AI?
This article provides company management with guidance on what needs to be considered when making decisions with AI.
The business judgment rule of Section 93 para. 1 sentence 2 AktG, the “safe harbor”, protects members of the Management Board and managing directors from personal liability if they could reasonably assume that they were acting in the best interests of the company on the basis of appropriate information when making a business decision.
Can AI replace expert advice?
If the company does not have sufficient specialist expertise of its own or the issue is too far-reaching, the management bodies must seek expert advice. However, the following legal requirements must be met before management can rely on expert advice:
- Specialist knowledge of the expert
- No specifications for the expert regarding the desired result
- Complete and transparent disclosure of the relevant facts
- Check whether all necessary and relevant information has been taken into account
- Consistent answers to obvious questions from non-specialist third parties.
If the requirements are met, the expert advice relieves the board members. However, if they consult the AI instead of a human, the requirements can only be used to a limited extent. AI systems do have analytical capabilities and an extensive database, especially when it comes to specific questions and the consideration of special features. However, AI results are not always error-free and it can happen that AI tools distort data, hallucinate or simply use unreliable sources. AI cannot be equated with human expertise and experience. It is therefore not enough for the above-mentioned requirements to be met – if this is possible at all.
An erroneous AI result is not comparable to an erroneous assessment by a technical expert. It can therefore not easily replace expert advice at present and would therefore not be able to relieve a management body of its personal liability.
To what extent can managing directors and board members rely on AI-generated results?
Managing directors and board members must base their decisions on sound information. AI systems can support the processing of information, but cannot replace human review.
Board members should therefore ensure that AI-generated results are not adopted unfiltered. Above all, each result must be checked with regard to its exact context and topicality. It is not uncommon for AI to use only fragments from sources to find a solution to the question and thus draw incorrect conclusions. The sources used by the AI must therefore be examined in detail, scrutinized and the process documented in order to comply with its own obligations.
AI can provide an initial assessment and therefore be a helpful tool in the decision-making process. However, it cannot completely relieve managing directors and board members of their duty to review and monitor.
If they rely on an AI-generated result without checking it, the management may be personally liable for any resulting damage.
How should the use of AI be documented?
It is important that the information on which the management’s decision was based can be traced afterwards.
AI tools often only store the results for a short period of time. It is therefore advisable to precisely document the use of AI and the careful review of the results used.
The documentation should include the following aspects, among others:
- Responsibility of the user: Name, date, compliance with internal approval processes, review of AI-generated results with regard to context, timeliness and sources
- AI tool: Name, version, date
- Prompts used and documents provided
- Purpose of use and the desired end product (evaluation, analysis, expert opinion, letter).
Furthermore, if possible, corresponding PDF files of the prompts and the results should be created.
If the management has not adequately documented the basis for the decision, it cannot prove in case of doubt that it has obtained appropriate information. It cannot invoke the business judgment rule.
Does the Business Judgement Rule require the use of artificial intelligence?
One exciting question is to what extent management can even be obliged to use AI. The Federal Court of Justice ruled in 2008: When making decisions, all available sources of information of a factual and legal nature must be exhausted. On this basis, managing directors and board members must carefully assess the advantages and disadvantages of the existing options for action(BGH, decision of July 14, 2008 – II ZR 202/07).
No one can evaluate information sources faster than artificial intelligence. What’s more, business decisions often have to be made at short notice. Managing directors and board members can use AI to consider information that would otherwise have taken several weeks to analyze and therefore could not have formed the basis for decision-making.
In practice, does this mean that you are even obliged to use AI tools in order to base decision-making on a larger volume of information or analyzed data?
There may be cases in which a management team also needs to use the AI integrated in its company in order to exploit all available information. This is especially true in cases where they also use AI for information gathering and analysis in their daily business operations. If there is no legal reason against the use of AI in such a case, such as confidentiality or data protection, the board member would probably not be protected under the Business Judgement Rule if they deliberately refrained from using AI and then made an incorrect decision. Personal liability would be a possibility, as the management would not have used all available sources of information. However, it is also important in this context that the AI-generated results and information are checked for plausibility.
You can also listen to our podcast on the topic of liability of board members.
