How many Monsters Inc movies are there

European Legal Framework for Artificial Intelligence (Part II)

Artificial intelligence
Reading time 7 min.

This is the continuation of the two-part contribution to the current developments in the field of regulation of artificial intelligence at European level. While Part I of the article discussed in particular the risk-based regulatory approach proposed by the European Commission, Part II analyzes the usability with regard to applications in legal transactions.


Technological applications in right-hand traffic

Right-hand traffic

The widely used (but hardly defined) term "legal traffic" is intended to generously describe all processes that are related to legal work processes. This does not only mean work steps in the context of legal services within the meaning of § 2 RDG.


Technological applications

The technologies used in legal traffic as understood in this way are diverse.1 For a better overview, a distinction should be made below between legal technology applications in the narrower sense and other applications.


Legal tech applications in the strict sense

That unites the words "Legal Services" and "Technology"2 The term “legal technology” (or “legal tech” for short) has not yet been defined by law.3 The judiciary is also reluctant to give concrete definitions.4 There are, however, a large number of suggestions in the literature.5 According to a broad approach, the term is intended to cover information technologies that are used in the legal field.6 From a practical point of view, legal technology is the sum of all automatically functioning applications that support or can take over legal activities.7


If one follows the application-related categorization of these legal tech applications already proposed elsewhere,8 the following case groups can be formed in particular:


Automated law enforcement: Applications whose aim is to determine and enforce user rights partially or fully automatically (assertion and defense of claims).


Automated document creation: Applications that automatically generate documents that can be used in legal transactions after users have provided the information required for this via an integrated query system.9 A distinction can be made between applications for generating one-sided and multiple-sided declarations of intent.


Legal process outsourcing: Describes the outsourcing of legal activities.


Document analysis: Applications that support users in their search for documents or texts with specific content.


Marketplaces and expert portals: Applications designed to help you find experts; often combined with other functions (general legal tips, evaluation options, etc.).


Other technological applications in right-hand traffic

There are also other technological applications that are used in right-hand traffic. This also includes office tech applications, i.e. office organization software such as time recording programs and document management,10 which are of course also used in non-legal fields of activity. This must be distinguished from legal research systems that enable systematic access to subject-specific sources and are specifically tailored to legal activities. The latter can also apply to chatbots.11


Other fields delimit the terms eJustice and eGovernment. While the eJustice initiative aims to promote the electronic exchange of documents between authorities and courts or lawyers, the eGovernment initiative aims to enable administrative procedures to be carried out electronically. The technological developments that are at least partially required for this are also the subject of EU funding initiatives.12


Technological applications and risk-based regulatory approach

The risk-based regulatory approach outlined in Part I of this article does not yet take into account AI-supported technological applications in legal transactions. Some of the applications presented here are likely to be referred to as AI systems, at least in the medium term in the sense of the definition developed by HEG KI.13 This would investigate whether these applications are associated with a high risk.


Legal market as a high-risk sector

The question of whether the legal market is a sector in which considerable risks are to be expected due to the nature of the typical activities cannot be answered in the negative without hesitation. Typical activities in the legal market can be associated with considerable risks, for example if existential issues are affected. This is also taken into account by the regulation of § 51 Paragraph 1 Clause 1 BRAO, which provides for professional liability insurance for lawyers admitted in Germany.14 This compulsory liability insurance15 serves in particular to protect clients from damage caused by legal activity.16 However, the diversity of legal activities means that less risky acts are carried out in the legal market. Evidence for this may be, for example, that only certain contracts require a standard notarization17 and the lack of which can result in the nullity of the legal transaction.18


Concrete use

It could therefore depend on the specific use of the AI-supported system in the legal market. With regard to legal tech applications, a differentiation could be made, for example, according to the categories proposed here. As an example, the technologically quite attractive category "Automated Document Creation"19 examined, it ultimately remains to be stated that the subject of the document to be created is important for the risk assessment.


Intermediate result

Although the intended risk-based regulatory approach for AI systems is fundamentally convincing, questions about general and specific risk assessment arise with regard to technological applications in legal transactions. Risk allocation becomes problematic, for example, in applications in which the specific use is always identical, but the risks associated with the use vary depending on the subject or situation being dealt with.


Possible solutions

The uncertainties with regard to risk assessment associated with the risk-based regulatory approach in general and with technological applications in legal transactions in particular can be countered with various approaches. They all have pros and cons.


RAPEX for AI systems?

Rapid alert systems have been established at European level to protect consumers from certain risks. The Rapid Alert System for Food and Feed (RASFF) and the Rapid Exchange of Information System (RAPEX) serve to quickly make risks transparent. The RAPEX system puts the probability of damage occurring in relation to the maximum possible severity of the damage. A risk index is derived from this. Based on the RAPEX system, information on risks from AI systems could be collected and reported. In this context, a comparative analysis of individual sectors and an assessment of the risks associated with typical legal transactions would be conceivable. With such an approach, however, there is the challenge of first having to create objective standards for risk assessment.


Regulatory sandbox for AI systems?

The term “regulatory sandbox” or “real laboratory” comes from information technology. Here, real laboratories were used in the course of software development to test new, potentially unsafe codes in a delimited framework in order to minimize the risk for the user when introducing new codes on the overall market.20 Sandbox solutions that have been established so far are united by the idea of ​​testing innovations under defined limits before launching them onto the market in order to identify potential risks and regulatory problems at an early stage and thereby protect consumers.21 A regulatory sandbox for AI systems22 could be used accordingly and enable the assessment of identifiable risks in individual sectors. According to the conclusions of the Council of the European Union on real-life laboratories and experimentation clauses of November 16, 2020, the legal-political will to increase the use of regulatory sandboxes is likely to be given at least in the beginning.23 However, regulatory sandboxes must be operated. Associated with this are questions about responsibilities and authorities.


Modification of the risk-based regulatory approach?

Alternatively, the risk categories set out in the White Paper could be more finely graded. For example, categories of medium risk levels could be considered, which are less based on an assignment to a specific sector and more on the specific use of an AI. Of course, this first and foremost raises the question of which legal framework should apply to the newly created “intermediate categories”.



The technologies that can be used in right-hand traffic, which is widely understood here, are diverse. This also includes so-called legal tech applications in the narrower sense. After an application-related categorization, different case groups can be formed, which can be useful in the legal evaluation of the applications.


At least some of the technological applications used in right-hand traffic are likely to be described as AI systems in the medium term in the sense of the definition developed by HEG KI. The intended risk-based regulatory approach for AI systems raises questions about general and specific risk assessment.


These uncertainties can be countered with various approaches. For example, a process based on the established RAPEX system could enable a comparative analysis of individual sectors and an assessment of the risks associated with typical legal transactions. A regulatory sandbox could have the same benefit for AI systems in general or for certain AI systems in particular. Last but not least, the risk categories could be graded more finely in order to enable a better assignment of individual AI systems. What unites the approaches is that they bring conceptual challenges with them. But it would be a shame not to discuss them for this reason.


1 For an overview, see for example Timmermann, Legal Tech Applications, 2020, pp. 101 ff.

2Breidenbach/Bald, REthinking Law 0/2018, 6, 6; Hartung, Thoughts on legal tech and digitization (in: Hartung/Bues/Half body, Legal Tech - The digitization of the legal market, 2018, margin no. 10) margin no. 17th

3 See the draft of a law for the promotion of consumer-friendly offers in the legal services market of the Federal Government of January 20, 2021, which does not require a definition of the term; but see BT-Dr. 19/9527, p. 1: "Legal tech applications, i. H. of algorithms to support and automate legal services“.

4 See, however, LG Berlin, decision of 07/03/2018 - 67 S 157/18: "Computer-based and standardized case analysis".

5 See, among others: Breidenbach/Bald, REthinking Law 0/2018, 6, 6; fiddler/Group, DB 2017, 1071, 1071; Frieze, NJW 2016, 2860, 2862 footnote 32; Reinemann, NJW supplement to issue 20/2017 - Innovations and Legal Tech For the 68th German Lawyers' Day in Essen, 6, 6; Hartung/Meising, NZFam 2019, 982, 983; Solmecke/Arends-Paltzer/Schmitt, What is Legal Tech - Application Error (in: Solmecke/Arends-Paltzer/Schmitt, Legal Tech - The digital transformation in the law firm, 2019, p. 25 ff.) P. 25; wagner, BB 2017, 898, 898.

6 Compare already wagner, BB 2017, 898; Wendt, Legal Tech - Get Automatic Right? (in: Ruppert/Hagen/Scrabble, Proceedings of the 5th Research Symposium “Science and Practice in Exchange on Current Challenges 2018”, 2018, p. 19) p. 22; see also Timmermann, Legal Tech Applications, 2020, p. 33.

7Wendt / Jung, ZIP 2020, 2209.

8 See. Wendt / Jung, ZIP 2020, 2209, also to the following.

9 See also Hartung/Meising, NZFam 2019, 982, 983.

10 See also on this Hartung, Thoughts on legal tech and digitization (in: Hartung/Bues/Half body, Legal Tech - The digitization of the legal market, 2018, margin no. 10) margin no. 21 f .; Timmermann, Legal Tech Applications, 2020, pp. 101 ff.

11Hartung, Legal Tech and professional law (in: Hartung/Bues/Half body, Legal Tech - The digitization of the legal market, 2018, margin no. 1031) margin no. 1042 ff.

12 See, for example, Call for proposals for action grants to support national or transnational projects: e-Justice (JUST-2021-EJUSTICE) from 2.3.2021.

13 Well, otherwise Werthmann, Part 3, § 22 Legal Tech (in: Boar/Heinze/Krügel/Steinrötter, Artificial Intelligence and Robotics, 2020, § 22 margin no. 1) margin no. 37, but see the notes on “Predictive Analytics” margin no. 39.

14 In addition, see also § 51a BRAO and § 59j BRAO.

15 See for example Staudinger / Halm / Wendt /Small, Insurance Law, III. Liability insurance law, AVB-RSV, § 1 margin no. 2.

16 Henssler / Prütting /Diller, § 51 BRAO margin no. 10 f.

17 E.g. those according to § 311b Abs. 3 BGB.

18 See § 125 sentence 1 BGB.

19 Well, otherwise Wormite, InTeR 2021, 22, 23.

20 See in detail on this J. Wendt, Corporate Finance No. 11-12 2020, 366, 367 f., With further references.

21 See. Rings / Ruof, Regulating Fintech in the EU 2020, p. 3; J. Wendt, Corporate Finance No. 11-12 2020, 366, 367.

22 See. Burchardt, Put the AI ​​in the sandpit !, DIE ZEIT from December 10th, 2020, p. 49.

23Council conclusions on real-life laboratories and experimentation clauses as instruments for an innovation-friendly, future-proof and resilient legal framework for dealing with disruptive challenges in the digital age of November 16, 2020.