Anonymous in Name Only: A Critical Review and a Reidentification Exercise on the European Commission's Anonymization Measures for Google Search Data under Article 6(11) of the DMA
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.
Nadia: We're looking at the paper: "Anonymous in Name Only".
Elias: Here's the gist of the paper. In this paper, we show why the anonymization measures proposed by the European Commission should not be treated as a sound anonymization practice because they rely on unjustified heuristics and fail to meet formal privacy guarantees.
Nadia: Critical Review of Anonymization Criteria The authors critically review the EC's technical and contractual measures, arguing that they fail to render the data anonymous and do not satisfy any of the criteria for effective anonymization set out by the European Data Protection Board (EDPB).
Elias: Specifically, they show that the EC’s approach is unrealistically optimistic, relies on heuristics rather than providing guarantees, and lacks technical justification.
Nadia: The proposed measures do not meet any of the three anonymization criteria established by the EDPB: no record isolation, no linkage, and no inference.
Paper summary: Elias: The paper starts by saying that because this is a concrete example of how regulators are trying to implement anonymization under the EU digital rulebook, it sets an important precedent for how other countries will handle data sharing later on.
Nadia: They look closely at the technical and contractual measures the EC proposed, specifically focusing on things like how they define residual risk and what their suppression thresholds are supposed to be.
Priya: I’m wondering if this review is just saying the EC missed some technical details, or if they’re saying the entire framework is fundamentally flawed from a privacy standpoint.
Elias: It's more than just missing details; they argue that the EC's approach relies on heuristics—basically educated guesses—rather than providing any concrete guarantees about privacy.
Nadia: They show that this reliance on heuristics lacks technical justification, which is a key part of their criticism against these proposed measures.
Priya: So if you can’t justify *why* the EC thinks they are safe, then we have to assume they aren't? That shifts the burden onto the regulators to prove safety rigorously.
Elias: Precisely. Furthermore, the paper shows that their risk assessment method is inappropriate because it doesn't provide a precise quantification of residual risk—they just don't define what "beyond residual risk" means in their guidelines or in GDPR itself.
Nadia: And another big technical flaw they highlight is the suppression mechanism itself; they show it only looks at each part of a search query separately, not whether putting those parts together makes the whole query unique enough to reidentify someone.
Priya: That sounds like a huge blind spot for privacy researchers because real-world identification often happens when you combine seemingly harmless pieces of data.
Elias: It’s definitely that kind of blind spot. The paper suggests this entire methodology is more similar to weaker data protection practices, like de-identification, which the GDPR doesn't actually allow if the data remains personal.
Conclusion: Nadia: Looking at the title, "Anonymous in Name Only," it really sums up their main point: these measures only give a superficial sense of privacy when you look closely at them under the DMA rules.
Elias: The authors are David Sánchez and Benet Manzanares and Josep Domingo-Ferrer from the ComSCIAM-Center for Computational Science and Applied Mathematics, so they bring a strong mathematical background to this critique.
Priya: What’s the bigger picture here for someone just listening? It seems like this paper is saying that relying on high-level policy language from a regulation isn't enough; you need actual technical proofs that the data stays anonymous.
Nadia: That’s right. They are essentially arguing that without formal privacy guarantees, these anonymization efforts don't satisfy the core principles of data protection law.
Elias: The implication is that if we want data sharing to happen securely under rules like the GDPR and DMA, we need a much stricter standard than what the EC has currently proposed.
Priya: So what does this mean for how we look at privacy-preserving techniques in general? It suggests that just applying a regulation doesn't automatically make the technical solution compliant.
Nadia: It means practitioners have to demand much stronger, more verifiable technical evidence before they trust that data is anonymized.
Elias: That’s the core message of this work on "Anonymous in Name Only: A Critical Review and a Reidentification Exercise on the European Commission's Anonymization Measures for Google Search Data under Article six(eleven) of the DMA" <ref:2611.17012#pg1,Anonymous in Name Only: A Critical Review and a Reidentification Exercise on>. It sets a high bar for what we consider 'safe' data handling.
Priya: For me, it’s about making sure that when we talk about data sharing, we aren't just checking boxes on a regulatory form and ignoring the actual technical risks involved in reidentification.
Department of Computer Science and Mathematics, ComSCIAM-Center for Computational Science and Applied Mathematics, CYBERCAT-Center for Cybersecurity Research of Catalonia · École normale supérieure, Université PSL
cs.CR, cs.IR
Submitted: 2026-11-12
Updated: 2026-11-12
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 83/100
The gist: In this paper, we show why the anonymization measures proposed by the European Commission should not be treated as a sound anonymization practice because they rely on unjustified heuristics and fail…
Key concepts
- Anonymization Criteria (EDPB)
- These are the standards set by the European Data Protection Board that data must meet to be considered truly anonymous. The paper argues the EC's measures fail because they do not satisfy these three core requirements: preventing record isolation, linkage, and inference.
- Heuristics vs. Guarantees
- Heuristics are rules of thumb or educated guesses used by the Commission instead of formal, mathematically proven methods. The authors contend that relying on heuristics is unrealistic because it does not provide concrete privacy guarantees; it lacks the necessary technical justification required for sound anonymization.
- Residual Risk Quantification
- This refers to precisely measuring the remaining privacy risk after anonymization. The paper claims the EC's risk assessment is flawed because it cannot define what 'beyond residual risk' means using existing guidelines, meaning the assessment of safety is imprecise.
Terminology
Summary
In this paper, we show why the anonymization measures proposed by the European Commission should not be treated as a sound anonymization practice because they rely on unjustified heuristics and fail to meet formal privacy guarantees. p. 0
Critical Review of Anonymization Criteria
The authors critically review the EC's technical and contractual measures, arguing that they fail to render the data anonymous and do not satisfy any of the criteria for effective anonymization set out by the European Data Protection Board (EDPB). Specifically, they show that the EC’s approach is unrealistically optimistic, relies on heuristics rather than providing guarantees, and lacks technical justification. p. 1 The proposed measures do not meet any of the three anonymization criteria established by the EDPB: no record isolation, no linkage, and no inference. p. 0
Flaws in Technical Measures and Risk Assessment
The analysis demonstrates that the EC’s risk assessment is inappropriate because it lacks a precise quantification of residual risk. The authors argue that the definition of beyond residual risk
cannot be found in the EDPB’s guidelines or GDPR. Furthermore, the suppression mechanism fails because it assesses each query component in isolation, failing to consider whether the combination of all components makes the query itself unique and thus potentially reidentifying. p. 7 The approach is criticized for being more reminiscent of weaker data protection practices such as de-identification, which cannot render personal data nonpersonal within the terms defined by the GDPR. p. 1
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved data anonymization must shift from heuristic methods to formal privacy models like k-anonymity to ensure that
no specific and meaningful inference can be drawn from the given data.
This means systems should implement ex ante privacy conditions rather than relying on post-hoc risk assessments. -
AI systems should incorporate a multi-layered defense against low-effort attacks by designing mechanisms that prevent reidentification through query linkage, such as ensuring that
search records can be linked not only to publicly accessible online sources such as LinkedIn and other social media platforms, but also to other search records belonging to the same user.
-
The system should move beyond simple suppression thresholds by adopting a utility-first approach where anonymization is
optimized the trade-off between privacy protection and utility preservation,
rather than suppressing queries based on arbitrary filters that lack formal privacy justification. -
Implement mechanisms that prevent attribute disclosure by ensuring that
no record isolation
andno linkage
criteria are met, specifically addressing how unique search records can be isolated even without background knowledge throughstraightforward and automatable attack patterns.
-
Develop robust methods to counter targeted attacks by automating the detection of specific identifiers like Eircodes or malformed email addresses, recognizing that attackers can
transform rare information that was likely to be suppressed into an observable signal that can be used to identify the corresponding records.
Sources
Related papers
- From Strategy Search to Policy Learning: MCTS-Guided Adversarial Co-Evolution for LLM Safety Alignment
- Privacy Leakage from Recurrent State in LLM Inference
- Incrementally Verifiable Oracle Aided Computation
- Privilege and confidentiality in generative AI workflows
- BadMoE: Backdooring Mixture-of-Experts LLMs via Routing-Aware Triggers and Task-Underutilized Experts
- CoVault: Secure, Scalable Analytics of Personal Data