Artificial Intelligence Annex

1. General provisions

1.1. This annex (hereinafter: “Annex”) constitutes the ARTIFICIAL INTELLIGENCE ANNEX pursuant to clause 2.14, clause 10.6 and Chapter 13 of the GENERAL TERMS AND CONDITIONS (hereinafter: “Terms”) governing the use of the Platform operated by WZZ Lab Informatikai és Szolgáltató Korlátolt Felelősségű Társaság (hereinafter: “Provider”). The Annex forms part of the Terms.

1.2. The Annex sets out the scope of the AI Services, the roles of the parties, the handling of input data and the Output, the rules on transparency and labelling, the framework for automated operation, and the procedure for changing models and providers. The processing of personal data is governed by the DATA PROCESSING AGREEMENT, and the accounting of AI Credits is governed by clause 9.15 of the Terms and Chapter 8 of the PLANS AND PRICING ANNEX; the Annex does not repeat these.

1.3. For the purposes of the Annex:

1.3.1. “Feature”: an AI Service specified in Chapter 2.

1.3.2. “Input”: data, text, image and instruction provided by the Merchant, the User or – in the case of the Feature for answering customer messages – the Customer, or used by the Feature from the Platform to perform the task.

1.3.3. “Output”: content or result generated by the AI Service on the basis of the Input. The conditions for using the Output are contained in Chapter 5 and clause 10.6 of the Terms.

1.3.4. “Experimental Feature”: a Feature made available on an experimental basis pursuant to clause 3.5 of the Terms.

2. Scope of the Features, their enabling and human involvement

2.1.1. AI Services include Features supporting the operation of the Store, the generation and management of content, as well as customer communication, in particular: the assessment of the fraud risk of orders (fraud screening); answering customer messages; the AI assistant available on the Administration Interface; the generation, supplementation and machine translation of textual and image content, as well as product data; the creation of appearance template code; as well as the generation of proposals and drafts relating to product feeds and categorisation. The list is not exhaustive. The scope of individual Features available from time to time and the details of their operation are contained in the Administration Interface and the public product documentation; the Annex sets out the contractual framework for the use of the Features.

2.1.2. The rules for using the Features are as follows. Unless provided otherwise, the Feature may be used upon enabling by the Merchant or the User, at their specific request, according to their permissions, by installing an associated application or in the Plans under the PLANS AND PRICING ANNEX; for certain Features, the Plan may establish usage limits — in particular, response and volume limits. An operating mode in which the Output is displayed or reaches the Customer without human approval (automatic mode) may exclusively be applied upon express enabling by the Merchant or the User. In deviation from this, the fraud screening Feature operates with a fraud screening application installed by default in the case of a new Store; its part applying artificial intelligence cannot be disabled separately, the application may be disabled in its entirety. The Feature for answering customer messages is disabled by default; the Merchant may select draft or automatic mode.

2.1.3. The rules on human involvement are as follows. Unless the operating rules of the given Feature provide otherwise, the Output qualifies as a proposal or draft, and saving, publishing or applying it requires the approval, saving or acceptance of the User; publishing and deleting files may occur exclusively by decision of the User. Financial operations and operations permanently deleting data may, in the case of all Features, be executed exclusively with the confirmation of the User. The Feature for translation saves and displays the completed translation without approval. In the automatic mode of the Feature for answering customer messages, the reply reaches the Customer without human approval; in the event of a request for human assistance or an error, the conversation is transferred to the User. The operation of the fraud screening Feature is governed by Chapter 8.

2.2. It is emphasised that the operation of the Features does not in all cases include human review; the manner of human oversight and involvement is set out exclusively in clause 2.1.3 and Chapter 8.

2.3. The Provider may introduce a new Feature within the scope under clause 2.1.1, and may develop the operation of existing Features without amending the Annex; it shall inform the Merchant of the new or modified Feature on the Administration Interface. The introduction of a new Feature shall not qualify as an amendment under clause 18.2 of the Terms if the use of the Feature is subject to separate enabling or request by the User; if the new Feature involves the processing of personal data, the regime of the DATA PROCESSING AGREEMENT shall govern. The phasing out of a Feature and material changes to its operation are governed by Chapter 10.

3. Role of the parties

3.1. With respect to the Features provided under its own name, the Provider qualifies as a provider within the meaning of the AI Act defined in clause 13.6 of the Terms, and the Merchant, as the party deploying the Feature towards Customers in its own Store, qualifies as a deployer within the meaning of the AI Act. Transparency obligations shall be governed by Chapter 7.

3.2. If the Merchant uses a Feature for a high-risk purpose under Annex III to the AI Act that differs from its intended purpose, or substantially modifies the Feature, the resulting obligations of a provider under Article 25 of the AI Act shall be borne by the Merchant. Use for a high-risk purpose is prohibited without the prior, separate written agreement of the Provider.

4. Input, data use and retention

4.1. The Merchant warrants that it is entitled to provide the Input and to process it with the Feature, and that the Input does not infringe any right of a third party. The processing of personal data contained in the Input shall be governed by the DATA PROCESSING AGREEMENT.

4.2. A task containing personal data may exclusively be directed to a model provider designated for this purpose in ANNEX III TO THE DATA PROCESSING AGREEMENT. The Provider shall not use the Merchant’s Input and the Output to train an artificial intelligence model, and shall only engage a model provider whose terms or the settings applied by the Provider exclude the use of the Input for training purposes. The training of the risk model under clause 4.4 shall not fall under this prohibition.

4.3. The Provider may use aggregated usage and quality data resulting from the operation of the Features, which do not contain personal data, for the development of the Service.

4.4. It is emphasised that, for the purpose of Platform-wide fraud and abuse prevention – by way of derogation from Chapter 11 of the Terms, in such a manner that the signal does not identify the submitting Merchant to other Merchants – the Provider processes, as an independent controller, the following signals linked to an IP address generated in the Stores: the Merchant’s manual approval following the assessment under the fraud screening Feature and the flagging of the order as fraud; the fraud warning of the payment service provider; the payment dispute and its closure; in the case of cash-on-delivery delivery, exclusively the verified refusal of acceptance; and the result of the automated anti-abuse check. The Provider logs the signals with the internal identifier of the submitting Store. From the signals, the Provider generates a statistical risk model, which it uses to evaluate orders of other Stores and the Platform’s automated anti-abuse check; the signal does not trigger a decision on its own in another Store, it is exclusively one element of the risk assessment. The key generated from the IP address may be linked to the signals for up to 365 days from the time of the underlying event, and the signal log may be retained for the same period. A signal originating from an event older than three hundred and sixty-five days may not be used. The 365 days is a uniform upper limit calculated from the time of the event, which is not extended by any delay in the receipt of the signal. If the Merchant’s manual fraud flagging proves to be erroneous, or the payment dispute closes in favour of the Customer, the Merchant shall rectify the signal, and the Provider shall withdraw or rectify the signal without undue delay and recalculate the risk assessment. The legal basis for the processing is the Provider’s legitimate interest in fraud prevention (Article 6(1)(f) of the GDPR); the transfer of the signals to the Provider is based on the Merchant’s legitimate interest in fraud prevention. The Merchant accepts this clause by separate acceptance under clause 5.3 of the Terms.

4.5. In its own privacy notice, the Merchant shall indicate the Provider as an independent controller, the purpose, legal basis and retention period under clause 4.4, as well as the availability of the Provider’s PRIVACY NOTICE. The Provider shall provide information on the processing under clause 4.4 in its PRIVACY NOTICE, and data subjects may contact it in this regard – including to object under Article 21 of the GDPR. The Merchant’s instruction for return or deletion under the DATA PROCESSING AGREEMENT – including Chapter 13 thereof – shall not, in accordance with clause 1.3 of the DATA PROCESSING AGREEMENT, extend to the signals under clause 4.4; these signals shall be deleted within the time limit set out in clause 4.4.

4.6. The Provider shall retain log and tracking data relating to the execution of the Features for up to 30 days; the accounting logs do not contain the full text of the instruction passed to the model. The itemised record of AI Credit deductions shall be governed by clause 9.15.3 of the Terms. Conversations of the Feature for answering customer messages and the Administration AI assistant shall be retained until the deletion of the Store or the deletion of the data subject’s data, respectively, and the assessment record of the fraud screening Feature shall be retained for 365 days from the assessment; thereafter, the Provider shall delete the risk data of the assessment record and pseudonymise its data relating to the order. In the event of termination of the contract, Chapter 13 of the DATA PROCESSING AGREEMENT shall apply. The assessment record shall not constitute a network signal under clause 4.4 and shall not extend the 365-day upper limit under clause 4.4; after the 365th day from the event, its data may not be used for network risk assessment under clause 4.4.

5. Output and intellectual property

5.1. Rights in respect of the Output that are held by the Provider and are transferable shall belong to the Merchant. Restrictions arising from the terms of the model provider shall also apply to the use of the Output.

5.2. Without prejudice to clause 15.6 of the Terms, it is emphasised that the Provider provides no intellectual property or other warranty of title in respect of the Output, and has no indemnification obligation in connection with the Output. The Provider does not answer for any infringing nature of the Output – including Output used by the Merchant without review.

6. Verification of Output

6.1. The Output is machine-generated and may contain errors, inaccuracies or outdated information. The Merchant shall substantively verify the Output prior to publication or use.

6.2. If the Merchant chooses an operating mode in which the Output appears publicly or reaches the Customer without human approval (in particular, the automatic mode of the Feature for answering customer messages, the operations of the Administration AI assistant without confirmation, and the Feature for translation), the Merchant shall be responsible for the consequences resulting therefrom. Enabling the automatic mode of the Feature for answering customer messages is subject to separate acceptance recorded on the Administration Interface.

7. Transparency and marking

7.1. In the automatic mode of the Feature for answering customer messages, the Provider shall ensure that the Customer is informed, at the latest at the time of the first response generated by artificial intelligence, that they are interacting with an artificial intelligence system.

7.2. The Provider shall mark in a machine-readable format the text Output generated by the Features that appears on the public interface of the Store. The Provider shall not mark the Output if the Feature performs a standard editing assistance task, or does not substantially alter the content provided by the Merchant – in particular, machine translation of the Merchant’s own text –, as well as text of a few words and content substantially rewritten by a human.

7.3. The Merchant shall not remove or obscure the marking and the information, and shall itself fulfil its deployer obligations under the AI Act – including informing Customers in its own privacy notice –.

8. Fraud screening

8.1. The fraud screening Feature assigns a risk score, a risk level and a statement of reasons to orders. The evaluation supports the Merchant’s decision; the Feature does not automatically reject or delete orders.

8.2. In the event of a low risk level, the system may automatically approve the order according to the configured thresholds; in the event of a medium risk level, the system automatically approves the order upon the expiry of the configured waiting time, in the absence of a decision to the contrary by the Merchant. In the event of a high risk level, the system holds the order pending human review, and the Merchant decides on it. If the Merchant does not make the decision within 24 hours of the evaluation – or, if the expiry of the card payment authorisation is earlier, within one hour prior to the expiry of the authorisation –, the Provider shall remind the Merchant at least every 24 hours until the decision. The expiry of the deadline without result does not in itself result in payment collection, the rejection or deletion of the order; the fate of the payment authorisation is governed by the rules of the payment service provider. The Merchant decides on the fulfilment of the order and any refund.

8.3. If the AI Credit available to the Merchant is exhausted, or the use of artificial intelligence is not possible for any other reason, the fraud screening Feature continues to operate without the use of artificial intelligence, with rule-based evaluation; in such cases, the evaluation may be based on fewer signals.

8.4. The automatic approval under clause 8.2 is based on the Merchant’s prior general setting; every other decision related to the order that entails adverse consequences for the Customer rests with the Merchant. The Merchant is responsible for informing Customers about the evaluation in its own privacy notice.

9. AI Credit

9.1. Features – with the exception of the limited operation of the fraud-filtering Feature under clause 8.3 – may not be used in the absence of AI Credit. The use of AI Credit shall be governed by clause 9.15 of the Terms and Chapter 8 of the PLANS AND PRICING ANNEX.

10. Change of model and provider, change in operation

10.1. The Provider may change the underlying model and provider route of the Feature within the scope under ANNEX III to the DATA PROCESSING AGREEMENT, without prejudice to the guarantees concerning the place of processing and the level of data protection, without prior notice, provided that the intended purpose and material operation of the Feature do not change adversely.

10.2. The engagement of a new Sub-processor or a new third country is governed by Chapter 7 of the DATA PROCESSING AGREEMENT.

10.3. The amendment procedure under clauses 5.6 and 18.2–18.4 of the Terms shall apply to the phasing out of a Feature, as well as to any change in the intended purpose or material operation of the Feature that is adverse to the Merchant. The prohibition under clause 4.2 may be lifted exclusively with the express acceptance of the Merchant and an amendment to the DATA PROCESSING AGREEMENT.

11. Prohibited use

11.1. The prohibitions of the ACCEPTABLE USE POLICY AND RESTRICTED PRODUCTS ANNEX shall apply to the use of the Features. In addition, the following in particular are prohibited:

a) any prohibited artificial intelligence practice under Article 5 of the AI Act;

b) circumvention of the security and content filters of the Features;

c) generation of pornographic or erotic content;

d) generation of false or misleading Customer reviews, or deceptive statements on health, financial or legal matters; and

e) deceptive concealment of the machine-generated origin of the Output.

12. Experimental Features

12.1. The Provider shall designate an Experimental Feature as such on the Administration Interface. Availability shall be governed by clause 2.2 of the AVAILABILITY AND SUPPORT ANNEX.

12.2. The Provider may modify or discontinue an Experimental Feature; it shall inform the Merchant of its discontinuation a reasonable time in advance, except where security or legal reasons make discontinuation immediately necessary.

12.3. The rules of this Annex on data use, retention and labelling shall apply unchanged to an Experimental Feature.

13. Order of precedence and amendment

13.1. In matters concerning an AI Service, in the event of any discrepancy between this Annex and the Terms, the order of precedence under clause 5.4 of the Terms shall apply; with regard to the processing of personal data, the DATA PROCESSING AGREEMENT shall take precedence.

13.2. The amendment of this Annex shall be governed by clauses 5.6 and 18.2–18.4 of the Terms.