2) Does the Use of Copyrighted Data to Train AI System fall within the scope of the Economic Rights monopoly, i.e. does it require the authorisation of the Copyright holder unless covered by an Exception? Please answer YES or NO. Please explain, e.g. if the authorisation is necessary only if acts of reproduction, representation, etc. are carried out, or in any circumstances (even if no act of reproduction, representation, communication, etc. has been carried out).

YES. In accordance with Art. 10(1) CopA, the author has the exclusive right to decide whether, when and how his or her work is used, regardless of the technology employed. As per Art. 10(2) CopA, the author has the right, in particular, to produce copies of the work (right of reproduction, lit. a). It is likely that the use of copyrighted data to train an AI System falls within the scope of this provision and therefore requires the authorisation of the copyright holder (unless covered by an Exception).

However, a more precise technical analysis is required to determine the stages of the input phase at which copies of protected works are made. Thus, although there is a consensus within Swiss legal doctrine that the storage of training data in a file or in a database constitutes a copy as per Art. 10(2)(a) CopA, this analysis has to be conducted with respect to the various stages that occur at the input level, which comprise data collection, data preprocessing, and the training of the AI model.

3) Are there any Exceptions that authorise the Use of Copyrighted Data to Train AI System without Copyright holder consent?

a. General Exceptions to Copyright, i.e. which are not special to the Use of Copyrighted Data to Train AI System (e.g. fair use, intermediate storage/temporary reproduction, etc.)? Please answer YES or NO. Please explain, e.g. the conditions of application of each Exception separately.

YES. There are several copyright exceptions that are not specific to the Use of Copyrighted Data to Train AI Systems.

For instance, Art. 19(1) CopA provides that: «Published works may be used for private use.» However, this provision applies to a limited number of cases, delineating three distinct meanings of the term «private use.»

Firstly, private use includes «any personal use of a work or use within a circle of persons closely connected to each other, such as relatives or friends» (lit. a). Consequently, the CopA allows a natural person to reproduce and, more generally, to use protected works for personal use or to share them with relatives or friends.

Secondly, private use further encompasses «any use of a work by a teacher and his/her class for educational purposes» (lit. b).

Finally, private use also means «the copying of a work in enterprises, public administrations, institutions, commissions and similar bodies for internal information or documentation.» Thus, copies of protected works can be circulated internally or stored on servers for access via the intranet, for example.

In addition, Art. 24a CopA states that: «The making of temporary copies of a work is permitted if:

• a. they are transient or incidental;
• b. they represent an integral and essential part of a technological process;
• c. their sole purpose is to enable a transmission of the work in a network between third parties by an intermediary or a lawful use of the work; and
• d. they have no independent economic significance.»

Therefore, those four cumulative conditions apply solely to transient or incidental copies that serve another purpose, such as browsing or caching, which speed up the downloading of data. Nevertheless, this provision does not extend to copies that are made to train AI models.

b. Special Exceptions to the Use of Copyrighted Data to Train AI System (e.g. Text and Data Mining -TDM Exception, etc.)? Please answer YES or NO. Please explain e.g. the conditions of application of each Exception separately.

NO. There are currently no copyright exceptions specifically designed for the copying of protected works to be used to Train AI Systems.

Article 24d CopA, which entered into force on 1 April 2020, applies to the use Its first paragraph provides that: «For the purposes of scientific research, it is permissible to reproduce a work if the copying is due to the use of a technical process and if the works to be copied can be lawfully accessed.» The Swiss Federal Council emphasises that this restriction in favour of scientific research is not limited to non-commercial research, and therefore also includes cooperations between universities and companies, which can lead to spin-off companies.

Furthermore, Art. 24d(2) CopA states that: «On conclusion of the scientific research, the copies made in accordance with this article may be retained for archiving and backup purposes.» Finally, Art. 24d(3) CopA excludes computer programs from its scope as follows: «This article does not apply to the copying of computer programs.»

4) Do the Exceptions provide for financial compensation for Copyright holders (e.g. a royalty paid to a collecting society, etc.)? Please answer YES or NO. Please explain.

NO. Article 19 CopA (Private use), Art. 24a CopA (Temporary copies) and Art. 24d CopA (Use of works for the purposes of scientific research) allow protected works to be used without any financial compensation.

However, Art. 20(2) CopA establishes a statutory license in the following circumstances: «Any person who reproduces works in any manner for private use under Article 19 paragraph 1 letter b [i.e., use by a teacher and his class for education purposes] or c [i.e., the copying of a work in enterprises, public administrations, institutions, commissions and similar bodies for internal information or documentation], or any person who does so as a third party under Article 19 paragraph 2 owes remuneration to the author.» Those claims can only be asserted by the authorised collective rights management organisation (Art. 20(4) CopA).

5) Can the author object to Use of Copyrighted Data to Train AI System on the basis of his/her Moral Rights (for example on the basis of the right to integrity, paternity, etc.)?

NO. An author enjoys various Moral Rights under CopA. Among those rights, the right to integrity as per Art. 11(1) and (2) CopA is the most likely basis for objecting to the Use of Copyrighted Data to train AI Systems.

Pursuant to the right to integrity, an author has, in principle, the discretionary right to decide whether, when, and how their work is modified (Art. 11(1)(a) CopA), used to create a derivative work, or incorporated into a collective work (Art. 11(1)(b) CopA). However, according to legal scholars, the mere technical transformation of a work, such as its incorporation into another medium or its recording using a different technology, does not qualify as a modification falling within the scope of the right to integrity. That is because the individual character of the work is not altered.

Accordingly, although it cannot be completely excluded that moral rights could entitle an author to object to the Use of Copyrighted Data to Train AI Systems depending on the circumstances, in particular on the degree of modification (e.g., reformatting, noise reduction) and whether a derivative work (e.g., translation) is created as part of the training, moral rights are more plausibly engaged at the output stage than at the input stage of AI training.

6) Does the Provider of an AI System have to make public the training Data used to train the AI System (transparency obligation)? Please answer YES or NO. If YES, please explain the degree of detail required (all works, categories according to sources – websites, etc.) and whether the developer/operator must keep the list of training Data or the training Data itself for a certain period of time?

NO. Under Swiss law, there is no explicit obligation to make public the training data used to train an AI System. Furthermore, there is no obligation on the part of the developer/operator to keep a list of training data or the training data itself for a certain period of time. However, transparency remains a key principle in the creation and training of AI systems as reflected in the Report to the Federal Council entitled «Overview of artificial intelligence regulation» which was published by the Federal Department of the Environment, Transport, Energy and Communication DETEC and Federal Office of Communications OFCOM on February 12, 2025.

7) Can the contents created (output) by an AI System be qualified as Copyright infringement in the following cases?

a. The output contains characteristic elements of one or more Copyrighted Data to Train AI System? Please answer YES or NO. Please explain if needed.

YES. If the output contains the characteristic elements of one or more copyrighted data, the contents created by an AI system can be qualified as copyright infringement.

b. The output is in the same style as one or more Copyrighted Data used to train AI System? Please answer YES or NO. Please explain if needed.

NO. Adopting only the style of copyrighted data, without the characteristic elements of one or more copyrighted data being recognisable, does not constitute an infringement of copyrights under current legal doctrine and practice.

c. The output is in all cases infringing if it has been trained with one or more infringed Copyrighted Data. Please answer YES or NO. Please explain if needed.

NO. Under current legal doctrine and practice, the output of an AI System that does not contain the characteristic elements (in their original or recognisably derived form) of one or more copyrighted data shall not be considered an infringement of copyrights merely because the AI System has been trained with one or more infringed copyrighted data.

d. In other cases? Please answer YES or NO. Please explain if needed.

NO.

8) Can the AI System itself be considered to infringe Copyright? Please answer YES or NO. If YES, please explain, e.g. under what conditions.

YES. But only if the AI System itself is considered as storing reproductions of copyrighted data. However, there is currently no consensus on the answer to this question, including from a technical standpoint.

9) Who is liable in case of Copyright infringement?

a. The Provider of an AI System? Please answer YES or NO. Please explain if needed.

YES. The Provider of an AI System who trained it in a way that infringes copyrights shall be liable for copyright infringement. However, it should be noted that merely making an AI system available does not automatically render the Provider liable for copyright infringement.

b. The user who exploits commercially the AI System? Please answer YES or NO. Please explain if needed.

YES. The user cannot be held liable for copyright infringement simply because the AI System has been trained in a way that infringes copyrights. However, the user may be held liable for copyright infringement if the output generated through an AI System infringes copyright or, depending on the circumstances, if the use of the input in connection with the AI System infringes copyrights.

c. The final user? Even if acting in good faith or unaware of the infringement? Please answer YES or NO. Please explain if needed.

YES. The final users cannot be held liable for copyright infringement simply because the AI System has been trained in a manner that infringes copyrights. However, the final users may be held liable if they use copyright infringing input or output. If the final users act in good faith, they may not be held liable for reparatory claims (damages), but they can still be subject to an injunction.

d. Any other person? Please answer YES or NO. Please explain if needed.

10) What sanctions can be imposed if it is found that Copyright has been infringed in order to train the AI System (because Copyrighted Data has been used with no authorisation)?

a. Injunction, destruction, etc. of Copyrighted Data used to Train AI System still present in the Data set or in the AI System? Please answer YES or NO. Please explain if needed.

YES.

b. Injunction, destruction, etc. of the AI System itself deemed to be infringing, because it has been trained with infringed Copyrighted Data? Please answer YES or NO. Please explain.

NO.

c. Injunction, destruction, recall from commercial channels, etc. of outputs deemed to be infringing? Please answer YES or NO. Please explain if needed.

YES.

d. The award of damages, including punitive damages? Please answer YES or NO. Please explain if needed.?

YES, but no punitive damages.

e. Confiscation of all or part of the profits generated by the operation of the AI System? Please answer YES or NO. Please explain if needed.

YES, insofar as these profits are generated by acts infringing copyright.

f. Any other sanctions? Please answer YES or NO. Please explain if needed.

NO.

11) Please explain, if and under what conditions outputs generated by an AI System are qualified as infringement of a Copyrighted work, e.g. because the output contains characteristic elements of the Copyrighted work, in the following cases:

a. In case the Use of the Copyrighted Data (work) to Train the AI System is covered by an Exception. Please answer YES or NO. Please explain if needed.

YES. The fact that copyrighted data has been lawfully used to train an AI System does not imply that the output generated by such systems may be lawfully used.

b. In case the Copyrighted work has NOT been used to train the AI System. Please answer YES or NO. Please explain if needed.

NO. Assuming that the copyrighted work was neither part of the prompt nor did the prompt refer to the copyrighted work in a way that would cause it to appear in the output.

In this case, the work is an independent parallel creation, i.e. a creation that was not made based on an existing work but only resembles one by coincidence. Since, according to the prevailing view in Switzerland, copyright does not have a «blocking effect», the owner of the copyrights in the existing work cannot intervene. However, in practice, there is a kind of factual presumption that if a younger work contains characteristic elements of an older work, this younger work was created based on the older work. As a result, it is de facto up to the author of the younger work to establish that he/she was not aware of the older work. In the context of AI output, such proof may regularly not be available, as the user does not know which works were used for the training and is therefore hardly able to present a defence.

c. In case the Use of the Copyright Data to Train AI System has been authorised, is the content generated (output) always licit, even if some for instance outputs contain characteristic elements of the Copyrighted work? Please answer YES or NO. Please explain if needed.

NO. See above answer to question 11) a).

12) Who is liable in case outputs infringes Copyright?

a. The Provider of an AI System? Please answer YES or NO. Please explain if needed.

NO. The Provider does generally not use the output work in a copyright sense. It will also hardly be possible to hold the Provider responsible as an instigator or co-perpetrator, since the Provider typically is unaware of the infringing acts of use. However, AI systems providers may face secondary liability if they contribute or aid infringement of a copyright in a sufficiently causal manner.

b. The user who exploits commercially the AI System? Please answer YES or NO. Please explain if needed.

NO. The user who exploits commercially the AI System would not be liable for the outputs, unless the commercial exploitation aims at infringing protected works.

c. The final user? Even if acting in good faith or unaware of the infringement? Please answer YES or NO. Please explain if needed.

YES. Final users also generally use the output work in a copyright sense. If final users act in good faith, they may not be liable for reparatory claims (damages), but they can still be subject to an injunction. Final users may also benefit from exceptions (e.g. private use).

d. Any other person? Please answer YES or NO. Please explain if needed.

As an independent programmer, if you code an AI System to specifically infringe Copyrights.

16) Should the Use of Copyrighted Data to Train AI System fall within the scope of the Economic Rights monopoly, i.e. should it require the authorisation of the Copyright holder as a matter of principle, unless covered by an Exception? Please answer YES or NO. Please explain, e.g. if the authorisation should be necessary only if acts of reproduction, representation, etc. are carried out, or in any circumstances.

YES. The Use of Copyrighted Data to Train AI should require the authorisation of the copyright holder as a matter of principle. Nevertheless, such authorisation should not be a prerequisite if it can be unequivocally demonstrated that the copyrighted data is solely utilised for the information it contains and not for any attributes protected under CopA.

17) Should there be Exceptions that allow the Use of Copyrighted Data to Train AI System without Copyright holder consent:

a. General Exceptions to Copyright, i.e. which are not special to the Use of Copyrighted Data to Train AI System (e.g. fair use, intermediate storage/temporary reproduction, etc.)? Please answer YES or NO. Please explain, e.g. the conditions of application of each Exception separately.

NO.

b. Special Exceptions to the Use of Copyrighted Data to Train AI System (e.g. TDM Exception, etc.)? Please answer YES or NO. Please explain e.g. the conditions of application of each Exception separately.

NO.

18) Should the Exceptions provide for financial compensation for Copyright holders (e.g. a royalty paid to a collecting society, etc.)? Please answer YES or NO. Please explain.

NO. But possibly licenses administered by collective management organisations.

19) Should the author be able to object to the Use of Copyrighted Data to Train AI System on the basis of his/her Moral Right (e.g. on the basis of the right to integrity, paternity, etc.)? Please answer YES or NO. Please explain if needed.

NO.

20) Should the Provider of an AI System be obliged to make public the training Data used to train the AI System (transparency obligation)? Please answer YES or NO. If YES, please explain, e.g. the degree of detail that should be required (all works, categories according to sources – websites, etc.) and whether the company should keep the list of training Data or the training Data itself for a certain period of time?

NO. The Provider of an AI System should not be obliged to make public the training data used to train the AI System. However, the Provider of an AI System should be required to keep a list of the datasets used to train the AI Systems for a certain period of time.

21) Should the contents created (output) by the AI System be qualified as Copyright infringement in the following cases?

a. The output contains characteristic elements of one or more Copyrighted Data to Train AI System? Please answer YES or NO. Please explain if needed.

YES.

b. The output is in the same style as one or more Copyrighted Data used to Train AI System? Please answer YES or NO. Please explain if needed.

NO.

c. The output is in all cases infringing if it has been trained with one or more infringed Copyrighted Data. Please answer YES or NO. Please explain if needed.

NO.

d. In other cases? Please answer YES or NO. Please explain if needed.

22) Should the AI System itself be considered a Copyright infringement? Please answer YES or NO. If YES, please explain, e.g. under which conditions.

NO. An AI System in itself should not be considered a copyright infringer. Whether its use constitutes copyright infringement should depend on the specific actions taken by the AI System, such as the unauthorised reproduction of copyrighted data during training or the generation of output containing recognisable copyrighted data without proper authorisation.

23) Who should be liable in case of Copyright infringement by the outputs?

a. The Provider of an AI System? Please answer YES or NO. Please explain if needed.

YES.

b. The user who exploits commercially the AI System? Please answer YES or NO. Please explain if needed.

c. The final user? Even if acting in good faith or unaware of the infringement? Please answer YES or NO. Please explain if needed.

d. Any other person? Please answer YES or NO. Please explain if needed.

24) What sanctions should be imposed if it is found that Copyright has been infringed in order to train the AI System, because Copyrighted Data has been used with no authorisation?

a. Injunction, destruction, etc. of Copyrighted Data used to train AI System still present in the Dataset or in the AI System? Please answer YES or NO. Please explain if needed.

YES.

b. Injunction, destruction, etc. of the AI System itself deemed to be infringing, because it has been trained with infringed Copyrighted Data? Please answer YES or NO. Please explain.

NO.

c. Injunction, destruction, recall from commercial channels, etc. of outputs found to be infringing? Please answer YES or NO. Please explain if needed.

YES.

d. Award of damages, including punitive damages? Please answer YES or NO. Please explain if needed.

YES.

e. Confiscation of all or part of the profits generated by the operation of the AI System? Please answer YES or NO. Please explain if needed.

YES. Although it could be challenging to determine which part is attributable to the infringed works.

f. Any other sanctions? Please answer YES or NO. Please explain if needed.

YES. Publication of judgements.

25) Please explain, in the following cases, if and under what conditions outputs generated by an AI System should be qualified as infringement of a Copyrighted work, e.g. because the output contains characteristic elements of the Copyrighted work, or is in the same style as the Copyrighted Work, etc.:

a. In case the use of the Copyrighted work to train the AI System is covered by an Exception. Please answer YES or NO. Please explain if needed.

NO. In case the use of the copyrighted work to train the AI System is covered by an exception, outputs should not be qualified as copyright infringement. However, there is currently no exception in the CopA that allows such use, so that any copy of a copyrighted work to train an AI System constitutes an infringement of the right of reproduction under Swiss copyright law (with the exception of the cases mentioned in question 3).

b. In case the Copyrighted work has NOT been used to train the AI System. Please answer YES or NO. Please explain if needed.

NO.

c. In case the Use of the Copyright Data to Train AI System has been authorized, is the content generated (output) always licit, even if some for instance outputs contain characteristic elements of the Copyrighted work? Please answer YES or NO. Please explain if needed.

NO. The answer to this question depends on the scope of the license that is granted to the AI company for the purpose of training its AI System. For instance, it is possible that a license may have been granted solely for the training of the AI System, without extending to the generation of new content by the AI system (output).

26) Who should be liable in case of Copyright infringement by the outputs?

a. The Provider of an AI System? Please answer YES or NO. Please explain if needed.

YES, if the AI System has been trained on copyrighted data without the authorisation of the authors or rights holders.

b. The user who exploits commercially the AI System? Please answer YES or NO. Please explain if needed.

NO. See answer to question 12) b).

c. The final user? Even if acting in good faith or unaware of the infringement? Please answer YES or NO. Please explain if needed.

YES. See answer to question 12) c).

d. Any other person? Please answer YES or NO. Please explain if needed.

NO.

27) Please comment on any additional issues concerning any aspect of the Use of Copyrighted Data to Train AI System you consider relevant to this Study Question.

28) Please indicate which industry sector views provided by in-house counsel are included in your Group’s answers to Part III.