Ina Ree­se Tax & Legal Prac­ti­ce – Flens­burg | Harrislee

AI liability: If the AI makes a mistake, the board of directors is still liable

AI liability: If the AI makes a mistake, the board of directors is still liable

Arti­fi­ci­al intel­li­gence ana­ly­ses mar­kets, pre­pa­res decis­i­ons and recom­mends acqui­si­ti­on tar­gets. Howe­ver, if the sys­tem makes a mista­ke and the com­pa­ny suf­fers signi­fi­cant dama­ge as a result, a cru­cial ques­ti­on ari­ses: who is lia­ble? The ans­wer is legal­ly clear-cut. Lia­bi­li­ty rests with the board of direc­tors or the mana­ging direc­tor, not with the machine.

The business judgement rule only provides protection where an independent assessment has been carried out

Under Ger­man com­pa­ny law, the so-cal­led ‘busi­ness jud­ge­ment rule’ appli­es. It sta­tes that mem­bers of the exe­cu­ti­ve board are not per­so­nal­ly lia­ble if they have taken a busi­ness decis­i­on in good faith, on the basis of ade­qua­te infor­ma­ti­on and in the best inte­rests of the com­pa­ny. It sounds simp­le, but it isn’t once AI comes into the picture.

The pro­blem is that AI pos­s­es­ses neither human jud­ge­ment nor busi­ness expe­ri­ence or intui­ti­on. Yet it is pre­cis­e­ly the­se ele­ments that form the basis of the busi­ness jud­ge­ment rule. Anyo­ne who sim­ply adopts a decis­i­on based on AI out­put wit­hout cri­ti­cal­ly scru­ti­ni­s­ing it and dra­wing on their own expe­ri­ence can­not invo­ke the pro­tec­tion affor­ded by this rule. The lia­bi­li­ty risk then remains in full.

AI is not an expert opinion

If board mem­bers or mana­ging direc­tors are unable to form their own jud­ge­ment on spe­ci­fic tech­ni­cal mat­ters, they may rely on exter­nal expert opi­ni­ons. Howe­ver, this pre­sup­po­ses that the experts are inde­pen­dent, have not been given any out­co­me-ori­en­ted ins­truc­tions, and are awa­re of all rele­vant infor­ma­ti­on. As things stand, AI can­not meet the­se requi­re­ments. It is pro­ne to errors, uses sources that are some­ti­mes non-trans­pa­rent, and lacks its own pro­fes­sio­nal jud­ge­ment. An erro­n­eous AI result the­r­e­fo­re does not reli­e­ve the company’s manage­ment of legal liability.

Documentation is crucial

Anyo­ne who incor­po­ra­tes AI into decis­i­on-making pro­ces­ses must docu­ment this careful­ly. In the event of a legal dis­pu­te, it is essen­ti­al that it is clear on what infor­ma­ti­on a decis­i­on was based. The fol­lo­wing points should be recorded:

  • Name of the per­son who used the AI and the date of use
  • AI tool used, inclu­ding ver­si­on and date
  • Prompts used and docu­ments uploaded
  • Pur­po­se of use and desi­red outcome
  • How the AI results were che­cked for time­line­ss, sources and context
  • Com­pli­ance with inter­nal appr­oval processes

As many AI tools only store results tem­po­r­a­ri­ly, it is advi­sa­ble to save prompts and respon­ses as PDF files.

The EU AI Act introduces additional obligations

The EU AI Act intro­du­ces fur­ther requi­re­ments for so-cal­led high-risk AI sys­tems: a con­ti­nuous risk manage­ment sys­tem, requi­re­ments regar­ding data qua­li­ty and human over­sight, as well as docu­men­ta­ti­on obli­ga­ti­ons. The so-cal­led duty of lega­li­ty obli­ges board mem­bers to ensu­re com­pli­ance with the­se requi­re­ments. Anyo­ne who igno­res this is in breach of their duties, with cor­re­spon­ding lia­bi­li­ty con­se­quen­ces. AI is the­r­e­fo­re not just a tech­no­lo­gi­cal issue, but also a com­pli­ance issue.

What this means for managing directors of limited liability companies (GmbHs)

The­se legal prin­ci­ples do not app­ly sole­ly to public limi­t­ed com­pa­nies. Mana­ging direc­tors of limi­t­ed lia­bi­li­ty com­pa­nies (GmbHs) are also obli­ged to exer­cise due care in their decis­i­on-making. Anyo­ne who adopts AI results wit­hout due con­side­ra­ti­on and ther­eby brea­ches their duty of care as a busi­ness owner is per­so­nal­ly lia­ble. Neither the IT depart­ment nor the legal depart­ment can deal with this mat­ter on the manage­ment body’s behalf. The respon­si­bi­li­ty lies with the manage­ment body its­elf. Anyo­ne who is unsu­re how to ensu­re that the use of AI within their own com­pa­ny is car­ri­ed out in a way that mini­mi­ses lia­bi­li­ty should seek legal advice at an ear­ly stage.

This post and the accom­pany­ing image were crea­ted with the help of AI and have been edi­to­ri­al­ly reviewed.

You might also be interested in this