EUTechRegResearch
MenuClose menu
Digital Omnibus legislative tracker
Texts to 22 Oct 2026

Digital Omnibus proposal· Recital 33

Analysis in our dossiers
AI development and residual sensitive data

Text

Recital 33 Digital Omnibus proposal as printed in the Council text of 22 October 2026

Source Council, 22 October 2026. Recital 33: ST 14756/26, page 31 · pp. 31–32 · exact text page

The development of certain AI systems and AI models may involve the collection of large amounts of data, which can include personal data and special categories thereof. Special categories of personal data may incidentally and residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing and the controller did not intend to process such personal data and has taken the appropriate technical and organisational measures to avoid such processing, such as the filtering out of special categories of personal data at the collection stage.

In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and erase special categories of personal data, including technical capabilities, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be limited to incidental and residual processing of special categories of data in the context of the development and technical operation of AI systems or AI models. The derogation should not be understood as covering the processing of special categories of personal data directly acquired by or provided to the AI system or AI model in the course of the deployment. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures throughout the entire lifecycle, that is to say during the development and operation, of an AI system or AI model and, once it identifies such data, effectively erase them. If erasure would prove technically impossible or require manifestly disproportionate effort, notably where the erasure of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, or would be technically impossible, the controller should effectively protect such data from being further processed or processed for other purposes, in particular being used to infer outputs, being disclosed or otherwise made available to third parties. In line with the accountability principle, the controller should document its assessment and have processes in place to monitor and demonstrate the effectiveness of these measures. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing for example if special categories of personal data relating to a natural person’s voice, image or other personal characteristics are used in a way that goes beyond incidental or residual existence. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679 or on other Union law, such as Regulation (EU) 2024/1689 regarding the processing of special categories of personal data for the purpose of ensuring bias detection and correction. The notion of AI system and AI model should be understood in the same manner as in Regulation (EU) 2024/1689.

How the recitals changed

Each text against the text before it

Recital Commission19 Nov 2025Rec 33Council21 May 2026Rec 33–33aCouncil10 Jun 2026Rec 33–33aCouncil18 Jun 2026Rec 33–33aCouncil3 Sep 2026Rec 33–33aCouncil2 Oct 2026Rec 33–33aCouncil22 Oct 2026Rec 33–33a
Compared withlaw in forcethe proposal21 May text10 Jun text18 Jun text3 Sep text2 Oct text
Recital 33 33 Added 33 Changed 33 Changed 33 Unchanged 33 Changed 33 Changed 33 Unchanged
Recital 33a In neither text 33a Added 33a Changed 33a Changed 33a Changed 33a Changed 33a Unchanged

Each text against the text before it: +Added in this text, not in the text before~Changed wording changed=Unchanged same as the text before— in neither text

Parliament amendments propose other recitals under these numbers: Recital 33a (new).

Parliament amendments

18 amendments to this recital

9 amendments would delete the whole recital: members of The Left, S&D, Greens/EFA and non-attached members, and the IMCO rapporteur.

By political group; co-sponsored amendments count for each group
TypeThe LeftS&DGreens/EFARenewEPPECRPfENIRapporteur
− Deletion213····21
~ Alternative1··13321·

By political group

  • − Deletion The Left 2 S&D 1 Greens/EFA 3 NI 2 Rapporteur 1
  • ~ Alternative The Left 1 Renew 1 EPP 3 ECR 3 PfE 2 NI 1

Show

  1. 20 Deletion Alex Agius Saliba (rapporteur)Rapporteur · IMCO Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Rapporteur
    Alex Agius Saliba
    Group
    Rapporteur
    Targets
    Digital Omnibus proposal Recital 33
    Document
    IMCO draft opinion
    Printed header
    Recital 33
  2. 120 Deletion Arash SaeidiThe Left · JURI Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Arash Saeidi
    Group
    The Left
    Targets
    Digital Omnibus proposal Recital 33
    Document
    JURI amendments 69–296 to the draft opinion
    Printed header
    Recital 33
  3. 121 Deletion David CormandGreens/EFA · JURI Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    David Cormand
    Group
    Greens/EFA
    Targets
    Digital Omnibus proposal Recital 33
    Document
    JURI amendments 69–296 to the draft opinion
    Printed header
    Recital 33
  4. 169 Deletion David CormandGreens/EFA · IMCO Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    David Cormand on behalf of the Verts/ALE Group
    Group
    Greens/EFA
    Targets
    Digital Omnibus proposal Recital 33
    Document
    IMCO amendments 125–328 to the draft opinion
    Printed header
    Recital 33
  5. 170 Deletion Kateřina KonečnáNI · IMCO Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Kateřina Konečná
    Group
    NI
    Targets
    Digital Omnibus proposal Recital 33
    Document
    IMCO amendments 125–328 to the draft opinion
    Printed header
    Recital 33
  6. 296 Deletion Sibylle Berg +1NI · ITRE–LIBE Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Sibylle Berg, Martin Sonneborn
    Group
    NI
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  7. 297 Deletion Markéta GregorováGreens/EFA · ITRE–LIBE Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Markéta Gregorová on behalf of the Verts/ALE Group
    Group
    Greens/EFA
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  8. 298 Deletion Marina Kaljurand +8S&D · ITRE–LIBE Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Marina Kaljurand, Elena Sancho Murillo, Brando Benifei, Birgit Sippel, Alex Agius Saliba, Francisco Assis, Elisabeth Grossmann, Kristian Vigenin, Matjaž Nemec
    Group
    S&D
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  9. 299 Deletion Pernando Barrena ArzaThe Left · ITRE–LIBE Deletes the recital

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Pernando Barrena Arza
    Group
    The Left
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  10. 300 Alternative Aura Salla +8EPP · ITRE–LIBE “for incidental and residual processing of special categories of data in the context of the development and technical operation of AI systems or AI…”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including which may include personal data and special categories thereof. Special categories of personal data may incidentally and residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing and while the controller did not intend to process such personal data and has taken the appropriate technical and organisational measures to avoid such processing, such as the filtering out of special categories of personal data at the collection stage. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, including technical capabilities derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed for incidental and residual processing of special categories of data in the context of the development and technical operation of AI systems or AI models. The derogation should not be understood as covering the processing of special categories of personal data directly acquired by or provided to the AI system or AI model in the course of the deployment. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle, such as the development and operation, of an AI system or AI model, and, once it identifies such data, effectively remove them. If removal would require manifestly disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being further processed or processed for other purposes, in particular used to infer outputs, being disclosed or otherwise made available to third parties. The controller should document its assessment and have processes in place to monitor and demonstrate the effectiveness of these measures. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing, for example if special categories of personal data relating to a natural person’s voice, image or other personal characteristics are used in a way that goes beyond incidental or residual existence. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679 or in another Union law, such as Regulation (EU) 2024/1689 regarding the processing of special categories of personal data for the purpose of ensuring bias detection and correction. The notion of AI system and AI model should be understood in the same manner as in Regulation (EU) 2024/1689.

    Tabled by
    Aura Salla, Niels Flemming Hansen, Ana Miguel Pedro, Paulo Cunha, Adina Vălean, Henrik Dahl, Andrea Wechsler, Pekka Toveri, Christian Ehler
    Group
    EPP
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
Show 8 more amendments
  1. 301 Alternative Diana Iovanovici ŞoşoacăNI · ITRE–LIBE “The derogation therefore aims to strike a balance between the need to develop and improve artificial intelligence systems and the fundamental…”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679. The derogation therefore aims to strike a balance between the need to develop and improve artificial intelligence systems and the fundamental obligation to protect data subjects’ rights. It is not a general exception to the protection system established by Regulation (EU) 2016/679, but rather a limited and conditional mechanism applicable solely in situations where the presence of data falling under special categories is incidental and unintended. Controllers remain responsible for implementing data protection principles from the design stage and by default (‘privacy by design’ and ‘privacy by default’), as well as for demonstrating compliance with their obligations under data protection legislation. The legal framework thus aims to facilitate innovation in the field of artificial intelligence without detracting from the high level of personal data protection guaranteed in the European Union and without violating data subjects’ fundamental rights and freedoms.

    Tabled by
    Diana Iovanovici Şoşoacă
    Group
    NI
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  2. 302 Alternative Irena Joveva +5Renew · ITRE–LIBE “Moreover, this derogation does not encompass data provided by, or generated through the activity of, end users of online platforms or core platform…”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed.In order to reconcile the technical realities of massive, unstructured data ingestion during the initial model training with the high level of protection required for sensitive data, a narrowly tailored derogation from the prohibition under Article 9(1) should be provided not to disproportionately hinder the development of AI. This derogation must be interpreted strictly and limited exclusively to entirely inadvertent data collection during the initial development phase, rather than creating operational privilege. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoidprevent the processing of those data, takes the appropriate measures during thetraining entire lifecycle of an AI system or AI modelphase and, once it identifies such data, effectively removeerase them.it Ifprior removal would require disproportionate effort, notably whereto the removalcompletion of specialthat categoriesphase, ofthereby preventing sensitive data memorisedfrom inbecoming structurally embedded within the AI systemsystems or AI modelmodels. would require re-engineeringConsidering the AIlimitations systemof ormachine AIunlearning, where a model has already completed training and subsequent erasure is proven to be technically impossible, the controller should effectively protect such underlining data from being used to infer outputs, be reverse-engineered, lead to re-identification of data subjects or being disclosed or otherwise made available to third parties. Additionally, in such cases where erasal of data is technically unfeasible the controller should fully document this technical impossibility and notify competent supervisory authority while deploying robust mitigating measures to secure underlining data. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679. Moreover, this derogation does not encompass data provided by, or generated through the activity of, end users of online platforms or core platform services and should therefore be strictly excluded from this derogation across all stages of a AI system or AI model’s lifecycle, ensuring that these systems or models cannot exploit user interactions or prompts to bypass protection under Article 9 of Regulation (EU) 2016/679.

    Tabled by
    Irena Joveva, Michael McNamara, Raquel García Hermida-Van Der Walle, Oihane Agirregoitia Martínez, Veronika Cifrová Ostrihoňová, Fabienne Keller
    Group
    Renew
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  3. 303 Alternative Julie Rechagneux +7ECR, PfE · ITRE–LIBE “Where special categories of personal data are collected incidentally and residually in the course of web scraping from publicly accessible sources…”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retainedextracted or inferred from, in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties.If removal would require disproportionate effort, notably where the removal of special categories of data inextractable from in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. Where special categories of personal data are collected incidentally and residually in the course of web scraping from publicly accessible sources, the prohibition laid down in Article 9(1) applies to the controller within the framework of its responsibilities, powers and capabilities, in accordance with the case-law of the Court of Justice (Case C-136/17, GC and Others), provided that the controller implements measures to prevent the collection and the dissemination of such data, such as applying filters, excluding sources structurally containing such data, deleting such data immediately upon identification, preventing their extraction from the model, and monitoring the outputs of the system. The mere accessibility of personal data online does not constitute consent, nor does the absence of a technical measure opposing scraping. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Justification

    Aligns the terminology with the technical reality of AI models and codifies the ‘responsibilities, powers and capabilities’ test established by the Court in GC and Others, as applied by the EDPB in its Guidelines 03/2026 on web scraping. This provides legal certainty for the incidental collection of special categories without widening the derogations of Article 9(2).

    Tabled by
    Julie Rechagneux, Mélanie Disdier, Aleksandar Nikolic, Pierre-Romain Thionnet, Alexandre Varaut, Fabrice Leggeri, Marion Maréchal, Christophe Bay
    Group
    ECR, PfE
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  4. 304 Alternative Sebastian Tynkkynen +1ECR · ITRE–LIBE “taking into account the technical feasibility, the available resources, the nature of the AI system or model, and state of the art capabilities”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the technical, organisational and state of the art capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in ana manner it considers to be effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, taking into account the technical feasibility, the available resources, the nature of the AI system or model, and state of the art capabilities, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering or significant retraining of the AI system or AI model, the controller should take appropriate measures proportionate to the risks posed that effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Tabled by
    Sebastian Tynkkynen, Diego Solier
    Group
    ECR
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  5. 305 Alternative Andrea Wechsler +5EPP · ITRE–LIBE “for the performance of a task carried out in the public interest or in the exercise of official authority vested in them, Article 6(1), point (e), in…”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof.The development, training, testing, validation, deployment, operation and monitoring of AI systems and AI models may require the processing of personal data. Such processing may constitute a legitimate interest of the controller or a third party within the meaning of Article 6(1), point (f), where appropriate, provided that the interests or fundamental rights and freedoms of the data subject do not override that interest, taking particular account of whether the data subject is a child. Where public authorities process personal data for the performance of a task carried out in the public interest or in the exercise of official authority vested in them, Article 6(1), point (e), in conjunction with Union or Member State law referred to in Article 6(3), should remain the appropriate legal basis. Controllers should use non-personal or anonymised data where the intended purpose can reasonably and effectively be achieved by using such data. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties.Where the processing of personal data is necessary, pseudonymised data should be used wherever the intended purpose can reasonably and effectively be achieved by using such data. Appropriate technical and organisational measures should be implemented throughout the lifecycle of the AI system or AI model. Such measures should include data minimisation during the selection of sources and the training, testing and validation of the system or model, the use of state-of-the-art privacy-enhancing technologies, and measures to prevent memorisation, regurgitation, data leakage and the unauthorised disclosure of personal data through outputs. Personal data should only be included in an output where this is necessary for the intended purpose, the recipient is authorised to receive those data and the disclosure has a legal basis under Article 6 and, where applicable, Article 9. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.Particular protection should be afforded to special categories of personal data referred to in Article 9(1).

    Tabled by
    Andrea Wechsler, Marie-Sophie Lanig, Stefan Köhler, Alexandra Mehnert, Verena Mertens, Sabine Verheyen
    Group
    EPP
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  6. 306 Alternative João OliveiraThe Left · ITRE–LIBE “it swiftly and effectively, regardless of the effort or cost required, unless the controller can objectively and manifestly demonstrate that it is…”

    (33) The development of certain AI systems and AI models mayfrequently involveinvolves the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed.In order to safeguard the protection of personal data, particularly data that falls within the definition of special categories, the controller should put in place, by design and by default, suitable, robust and demonstrably effective organisational and technical measures to prevent the collection of special categories of personal data, or any other form of processing thereof, in the context of training, testing or validation data sets. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, once it identifies such data, effectively remove them.The controller should take the appropriate measures during the entire lifecycle of an AI system or AI model and, without prejudice to the foregoing, where it identifies such data, should remove it swiftly and effectively, regardless of the effort or cost required, unless the controller can objectively and manifestly demonstrate that it is materially impossible to remove the data concerned, and not simply disproportionate or burdensome to do so. If removal would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.If the data cannot be deleted, the controller should take measures to prevent such data from being used to infer outputs, being disclosed or otherwise being made available to third parties, and should devise those measures in such a way as to enable the competent supervisory authority provided by Regulation (EU) 2016/679 to verify compliance therewith.

    Tabled by
    João Oliveira
    Group
    The Left
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  7. 307 Alternative Jana Nagyová +3PfE · ITRE–LIBE “and economically feasible measures to cease processing”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoidmitigate the risk of the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, onceis itnotified of or itself identifies such data, effectivelyimplement removetechnically themand economically feasible measures to cease processing. If removalcessation of processing would require disproportionate effort, notably where the removal of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Justification

    This amendment ensures interpretive consistency between the recital and the operative provision and prevents supervisory authorities or courts from using the unchanged recital as a basis for a more restrictive interpretation of the derogation than that intended by the amended Article 9(5). It also reflects the extended scope of the controller’s duty to act, which now covers situations in which the controller is notified of the presence of special categories of personal data by third parties, and not only cases of self-identification.

    Tabled by
    Jana Nagyová, Ondřej Knotek, Tomáš Kubín, Jaroslav Bžoch
    Group
    PfE
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33
  8. 308 Alternative Zala Černilec Tomašič +8EPP, ECR · ITRE–LIBE “and economically feasible measures to cease processing”

    (33) The development of certain AI systems and AI models may involve the collection of large amounts of data, including personal data and special categories thereof. Special categories of personal data may residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing. In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and remove special categories of personal data, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be allowed. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoidmitigate the risk of the processing of those data, takes the appropriate measures during the entire lifecycle of an AI system or AI model and, onceis itnotified of or itself identifies such data, effectivelyimplement removetechnically themand economically feasible measures to cease processing. If removalcessation of processing would require disproportionate effort, notably where the removalcessation of processing of special categories of data memorised in the AI system or AI model would require re-engineeringreengineering the AI system or AI model, the controller should effectively protect such data from being used to infer outputs, being disclosed or otherwise made available to third parties. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679.

    Justification

    This amendment aligns Recital (33) with the revised Article 9(5) by replacing obligations to ‘avoid’ and ‘remove’ special categories of personal data with proportionate duties to mitigate risks and cease processing through technically and economically feasible measures. It ensures consistency between the recital and the operative text, prevents overly restrictive interpretation, and clarifies that controllers must act when such data is identified or reported by third parties.

    Tabled by
    Zala Černilec Tomašič, Jan Farský, Ondřej Krutílek, Tomáš Zdechovský, Michał Wawrykiewicz, Henrik Dahl, Alexandr Vondra, Veronika Vrecionová, Lukas Mandl
    Group
    EPP, ECR
    Targets
    Digital Omnibus proposal Recital 33
    Document
    ITRE–LIBE amendments 251–400 to the draft report
    Printed header
    Recital 33