The EDPB’s Latest Word on Legitimate Interest: Five Changes Worth Reading Closely
On 10 November 2026 the European Data Protection Board adopted version 2.0 of its Guidelines 1/2024 on processing based on Article 6(1)(f) GDPR. The three-step test survives intact, but five changes will matter to anyone advising on advertising, children’s data, safeguards and AI.

Two years after the consultation draft, the EDPB has finalised its guidance on the legal basis that controllers invoke most often and that regulators find hardest to predict. Readers expecting a rewrite will be disappointed. The Board kept the cumulative test of a legitimate interest, necessity and balancing, and declined most of what industry asked for during the consultation.[1] Version 2.0 retains both the tracking passage and the strict necessity standard. What changed lies in the detail, and much of that detail reflects judgments delivered by the Court of Justice after October 2024.
Below are the five changes I consider most consequential, each assessed for its effect on legal certainty, its fit with existing case law and its impact on innovation in the EU technology and AI sector. All references are to paragraphs and pages of version 2.0.
1. Personalised advertising is direct marketing, and tracking will rarely survive the balance
The Board notes that “CJEU case-law suggests that personalised advertising may be equated with direct marketing” (para 109, p. 33). For processing of personal data for marketing purposes that involves “tracking individuals across multiple websites, locations, devices or services”, it concludes that “the balancing test would be very unlikely to yield positive results” (para 119, p. 36). The right to object to processing for direct marketing under Article 21(2) GDPR “is unconditional and irrespective of the legal basis relied on by the controller” (para 121, p. 37). Summarising the Court’s ruling in Mousse (C-394/23), the Board adds that the GDPR requires the controller to “inform the data subjects of the legitimate interest pursued at the time when the data is collected” (footnote 94, p. 23).
The upside. This is the clearest statement yet of where the line sits, and it builds on Meta Platforms v Bundeskartellamt (C-252/21) and on the transparency standard set in Mousse (C-394/23) rather than inventing new doctrine. In my view, it also favours contextual and less intrusive advertising models. The concern is not new: Shoshana Zuboff of Harvard Business School has described surveillance capitalism as a model that “unilaterally claims human experience as free raw material for translation into behavioral data” (The Age of Surveillance Capitalism, Profile Books 2019, p. 8). The guidelines respond to that concern without prohibiting personalised advertising as such.
The downside. “Very unlikely” is a presumption without a test. The guidelines give no threshold of volume, duration or data combination at which ordinary first-party personalisation becomes tracking, and the equation of personalised advertising with direct marketing rests on what the case law “suggests” rather than on a definitive holding. The practical effect may also favour gatekeepers that can secure consent at scale, while European publishers and smaller ad-funded services lose a fallback basis they cannot easily replace.
2. The soft opt-in now stands on its own
Following Inteligo Media (C-654/23), the Board accepts that where the conditions of Article 13(2) of the ePrivacy Directive are met for the sending of a direct marketing communication, “it is sufficient in itself as a legal basis and no legal basis under Article 6 GDPR” needs to be identified for that specific activity (para 115, p. 35).
The upside. This is genuine legal certainty. It removes a duplicated assessment that many authorities had demanded and lowers compliance costs for e-commerce, where existing-customer marketing is routine.
The downside. The ruling removes one layer of analysis but leaves the uncertainty in the conditions of Article 13(2) itself, in particular what counts as “similar products or services”, and those conditions are transposed differently across Member States in the absence of an ePrivacy Regulation. The relief is also narrower than it first appears. The Board limits sufficiency to “the specific activity of sending the communication” (para 115, p. 35). Everything around it, such as segmentation, profiling to select recipients or tracking whether messages are opened, still needs its own Article 6 basis and falls back into the balancing test, where the Board considers tracking across “websites, locations, devices or services” “very unlikely to yield positive results” (para 119, p. 36). Marketing teams that read the soft opt-in as a general licence for data-driven email campaigns will be disappointed.
3. Children: commercial interests yield
Where a controller’s interests conflict with the interests or fundamental rights and freedoms of a child, “the interests or fundamental rights and freedoms of the child should in general prevail” (para 95, p. 30). The Board considers that operations such as “profiling for marketing purposes and targeted advertising activities” generally do not fit the specific protection children are owed, subject to limited exceptions (para 95, p. 30). It refers to Article 28(2) DSA, which prohibits online platforms from targeted advertising based on profiling of minors (para 95, p. 31), and to General Comment No. 25 of the UN Committee on the Rights of the Child: “States parties should prohibit by law the profiling or targeting of children of any age for commercial purposes” (footnote 149, p. 31).
The upside. The position aligns the GDPR with the DSA and gives platforms a single, predictable standard across both regimes.
The downside. The guidelines do not say how a controller is to know it is processing a child’s data. Age assurance is itself processing, often of more intrusive data, and nothing in the text addresses that tension. The limited exceptions are not identified, which leaves edtech and gaming providers guessing. Importing a UN statement addressed to States into a controller-level balancing test also stretches its normative status.
4. Mitigating measures: only extra effort counts
“The impact weighed in the balancing test should therefore already be the minimised impact” (para 45, p. 18), that is, the impact remaining once the controller has complied with its GDPR obligations. Where the balance still tilts against the controller, it “may consider introducing additional mitigating measures going beyond what is legally required” (para 53, p. 21), and if it does so it “should be able to demonstrate how this changes the outcome of the balancing test” (para 56, p. 22). The examples include “Proposing an unconditional ‘opt-out’ from the outset”, enabling erasure “even when the specific grounds listed in Article 17(1) GDPR do not apply” and “Providing a reasonable period of time between the collection of personal data and their use” (para 54, p. 21).
The upside. This closes a circular argument in which ordinary compliance was presented as a safeguard. It makes legitimate interest assessments auditable and gives engineers a concrete menu. Orla Lynskey, then at LSE, warned that a broad interpretation of necessity “will do little to incentivize a change in current data processing practices” (Delivering Data Protection: The Next Chapter (2020) 21 German Law Journal 80, 84). By keeping the strict necessity standard and expecting safeguards beyond the legal minimum, the Board has chosen the narrow reading.
The downside. The guidelines give no indication of how much weight any individual measure carries, so each controller must build its own methodology for showing that a measure has changed the result, knowing that a supervisory authority may later reject it. That burden is conditional, since it arises only where controllers “implement such additional mitigating measures” (para 56, p. 22), but it is not proportionate: the accountability obligation in Article 5(2) GDPR applies to a start-up exactly as it does to a multinational. There is also a risk that the examples in paragraph 54, such as “Proposing an unconditional ‘opt-out’ from the outset” (para 54, p. 21), harden in practice into a baseline expectation. If supervisory authorities come to treat them as the norm rather than as measures “going beyond what is legally required” (para 53, p. 21), the bar for relying on legitimate interest will rise without any change in the text of the GDPR.
5. AI: available, but only for interests that are present and effective
For the development and deployment of AI models, the Board points to the “general considerations to take into account” set out in Opinion 28/2024 and gives as an example “developing the service of a conversational agent to assist users” (para 90, p. 29). It also refers to its Guidelines 03/2026 on web scraping (footnote 137, p. 29). The interest itself must meet the criterion that “The interest is present and effective and not hypothetical at the date of the processing” (para 16, p. 9). The Board further reports the Court’s finding in NTH Haustechnik (C-484/24) concerning a controller that “knew, or ought to have known, that certain personal data transmitted to them had been collected or stored unlawfully” (footnote 30, p. 9).
The upside. The Board confirms that legitimate interest is a viable basis for AI. That matters because consent is unworkable at training scale, and it keeps the guidelines consistent with Opinion 28/2024.
The downside. An interest that must be present and not hypothetical sits uneasily with general-purpose models whose downstream uses are by definition undetermined, so model training risks failing at the first step. The “ought to have known” standard imposes dataset provenance diligence that EU start-ups buying third-party datasets will struggle to meet, while incumbents training on their own data are untouched. The guidance is also spread across further documents, including the Guidelines 02/2026 on Anonymisation (footnote 72, p. 18), which makes it harder to apply.
[1] For example, FEDMA asked that the statement on intrusive profiling and tracking “should be removed from the draft guidelines” and that certain activities “benefit from a presumption of being legitimate”, while IAB Europe objected to “adding a requirement of strict necessity to the processing”; see FEDMA’s submission, p. 3, and IAB Europe’s submission, sections 1 and 3(a), EDPB public consultation on Guidelines 1/2024, refs 10/2024-0044 and 10/2024-0025)
Conclusion
Conclusion
The takeaways
Legal certainty improves where the Board follows the Court of Justice, as with Mousse, Inteligo Media and NTH Haustechnik. It weakens where the text relies on open-ended formulations such as “very unlikely”, “in general” and limited, unspecified exceptions. In practice, controllers should revisit their assessments for advertising, children’s data and AI training, record mitigating measures with a clear before-and-after rationale, verify the provenance of any third-party dataset and state the legitimate interest pursued at the point of collection.
Article provided by INPLP member: Stefan Martinić, LL.M. Harvard Law School, LL.M. LSE, CIPP/E (Law Office Stefan Martinić, Croatia)
Article provided by INPLP member: Stefan Martinić (Law Office Stefan Martinić, Croatia)
