The Wikimedia Foundation is publishing data on requests received from European Union countries during the covered period. In August 2023, the EU Digital Services Act (DSA) began applying to Wikipedia, after it was designated as a Very Large Online Platform

Regarding the structure of the projects and how they are moderated

As per our consistent practice for transparency reporting and the DSA’s requirements, this report highlights the difference between moderation actions taken by the Foundation and those taken by the community of volunteer contributors who create and curate the information on Wikimedia projects. Together, the community of nearly 260,000 volunteers compile and share information on notable subjects, citing reliable sources such as newspaper articles and peer-reviewed journals, according to the encyclopedia’s editorial policies and guidelines

On a project such as Wikipedia, the community conducts almost all the content moderation. Volunteers vigilantly defend against information that does not meet the site’s policies. They regularly review a feed of real-time edits to quickly address problematic changes; bots spot and revert many common forms of negative behavior on the site; and volunteer administrators (trusted Wikipedia volunteers with advanced permissions to protect Wikipedia) further investigate and address negative behavior. Administrators can take disciplinary action and block user accounts or IP addresses from further editing when there are repeated violations of Wikipedia policies, vandalism, undisclosed paid editing, or edit warring. This whole process of content moderation by Wikipedia volunteers is open and transparent. Everything from the way an article grows and evolves over time, to the citations used to verify the facts, to the content discussions amongst editors, are all publicly available on the article’s history and talk pages. Readers wishing to gain a broader appreciation of moderation activity across the projects more generally are invited to browse the many data interfaces available for this purpose, such as Wikistats, and the Logs function for each project (e.g. the Logs for English Wikipedia).

Meanwhile, though the community of volunteers address the vast majority of content and conduct issues on Wikimedia projects, the Foundation has a Trust and Safety team within its Legal Department that alerts and supports the volunteer community on some extreme cases when they are reported. In some instances, the Foundation investigates larger issues of disinformation and other systematic forms of disruptive behavior that might appear on the site. Where necessary, the Foundation may also take an Office Action in a small number of cases, such as when the behavior of contributors on Wikimedia projects threatens the safety of other editors, keeps false information on the platform, or prevents volunteer communities from functioning properly. These actions have been disclosed by the Foundation in its transparency reporting.

Regarding countries of origin

The Wikimedia Foundation is committed to protecting the privacy of all users of all Wikimedia projects: readers and editors alike. As such, we deliberately collect very little user data, practicing the principles of data minimization. The right to privacy is accordingly embedded into the Wikimedia projects and is at the core of how communities contribute to them. We protect user data and foster free knowledge by allowing users to partake in knowledge sharing activities across our platforms without providing any personal details, including their country of origin. In the vast majority of requests we receive to alter content or hand over user data, we do not know the country in which the requester is located or of which they hold citizenship. This report represents a best effort to disclose requests in which this information was obvious or could be reasonably deduced from the content of the request (such as the requester citing a country-specific law).

Regarding volume of requests

The Foundation always receives a low volume of requests. Because of this, many of the following sections in this report indicate no relevant requests under the DSA transparency reporting categories. Notably, many of the categories are inapplicable by virtue of the content featured on the projects. For example, because the Wikimedia projects do not contain advertisements, it is less likely that the Foundation will receive DSA requests related to “false advertisements.”  We anticipate that future reports will continue with this trend and will likely show low but non-zero numbers for the DSA categories where we do receive requests. 

In order to demonstrate the volume and range of topics usually brought to our attention, where possible, the Foundation reports the requests that fall within the DSA categories. The Foundation further notes that, despite our efforts to solicit more context, not all of the requests received provide sufficient details to enable clear categorization or to enable further action without more information from the requester. This is one such reality the DSA tries to address by allowing platforms to add qualitative descriptions alongside the quantitative tables to provide more context, something the Foundation has endeavored to present in its transparency reporting since it began publishing.

Orders from EU Member States

Orders to act against illegal content (DSA Art. 9 in relation to Art. 15(a))

During the covered period, the Foundation did not receive orders from Member States’ national judicial or administrative authorities that meet the necessary conditions to fall under Article 9 of the DSA. This Article requires platforms to report the number of orders to act against one or more specific items of illegal content under EU law or Member State law. To be transparent, the Foundation reports the following incident(s) of interest and their status. 

Country                                 Category                                                                                    Number                                  Number of specific items of informationStatus                                                                                  
GermanyCategory not specified under DSA; Accuracy of a Wikipedia article11Not an actionable order; Matter pending before authority

Requests, including informal communications from EU-based government bodies or representatives, for alteration & takedown are found in the Transparency Report.

Orders to provide information (DSA Art. 10 in relation to Art. 15(a))

During the covered period, the Foundation did not receive orders from Member States’ national judicial or administrative authorities that meet the necessary conditions to fall under Article 10 of the DSA. This Article requires platforms to report the number of orders directing them to provide information about specific individual recipients of the service under EU law or Member State law. To be transparent, the Foundation reports that it has received the following requests from local authorities.

Country                                 Category                                                                                        Number                                  Number of specific items of informationStatus                                                                                  
BelgiumInvestigations related to defamation11Responded, no disclosure
FranceCategory not specified under DSA; request not about a Wikimedia project

Not clear from request
1





1
Not a Wikimedia project



3
Responded, no disclosure



Responded, no disclosure
GreeceNot clear from request11Responded, no disclosure

Requests, including informal communications from EU-based government bodies or representatives, to provide user data are found in the Transparency Report.

Notice and action submissions (DSA Art. 16 in relation to Art. 15(b))

*48 Notices 0 Notices from trusted flaggers 0 Notices that resulted in an Office Action
2 Number of actions taken on the basis of law 37 Number of actions taken on the basis of terms & conditions
9 Pending resolution 35 days Median time to respond

During the covered period, the Foundation received the following notices containing allegations of illegal content on the projects. There were 0 notice(s) from trusted flaggers during this period.

For all Wikimedia projects, the community oversees and manages the bulk of content moderation activities. In many cases, the Foundation refers requests we receive to the community and defers to community decisions about project content. When the community asks the Foundation for input, we sometimes provide guidance on relevant laws or project policies. These notices are addressed by the community independently. For this reason, there are only very few notices that are resolved through the Foundation’s intervention. 

*There are a small number of notices that were brought to our attention through official mechanisms, though they are more appropriately for and actioned by the community. These notices are those sent through official mechanisms and meet the criteria listed under Article 16 of the DSA. The criteria, among others, require the sender to include a sufficiently substantiated explanation of the reason why specified information on the Wikimedia project is alleged to be illegal under EU or Member State law. We report these notices sent to the Foundation including, generally, the actions taken by the community (based on law or terms & conditions) to be more transparent about the numbers, although these notices do not trigger Office Action. If a notice submitted under Article 16 of the DSA resulted in the Foundation taking an Office Action, the number of notices, actions, and their bases shall be labelled clearly in this section. 

In arriving at the median time, we consider when a notice was received through an official mechanism and the time it took to respond to the sender, noting how the projects work and any possible next steps available. Requests that do not meet the DSA criteria for whatever reason are included in the general Transparency Report under alterations & takedowns.

ProjectCountryAlleged ClaimNotices        
Commons              Belgium         Data protection and privacy violations: Missing processing ground1
DenmarkData protection and privacy violations: Missing processing ground1
FranceData protection and privacy violations: Missing processing ground

Intellectual property infringement: Copyright infringement
1



1
Germany       Data protection and privacy violations: Missing processing ground

Intellectual property infringement: Copyright infringement
2



1
Netherlands            Cyber violence: Harassment

Illegal or harmful speech: Defamation
1

1
SpainData protection and privacy violations: Missing processing ground

Intellectual property infringement: Copyright infringement
2



1
WikidataItalyData protection and privacy violations: Right to  be forgotten1
WikipediaAustriaData protection and privacy violations: Right to be forgotten1
BelgiumIllegal or harmful speech: Defamation1
FranceData protection and privacy violations: Missing processing ground

Illegal or harmful speech: Defamation

Intellectual property infringement: Trademark infringement
2


2


1
GermanyConsumer information: Misleading info goods & services

Cyber violence: Harassment

Data protection and privacy violations: Missing processing ground

Data protection and privacy violations: Right to be forgotten

Illegal or harmful speech: Defamation

Illegal or harmful speech: Other misinformation

Intellectual property infringement: Copyright infringement

Intellectual property infringement: Trademark infringement
1


1

1


2


2

4


1



2
HungaryData protection and privacy violations: Missing processing ground1
ItalyCyber violence: Harassment

Data protection and privacy violations: Right to be forgotten

Illegal or harmful speech: Defamation
1


4

1
PolandData protection and privacy violations: Missing processing ground

Illegal or harmful speech: Other misinformation
1



1
SpainData protection and privacy violations: Missing processing ground

Illegal or harmful speech: Defamation
3


1
WiktionaryFranceData protection and privacy violations: Right to be forgotten1

Content moderation by the Wikimedia Foundation (DSA Art. 15(c))

The following section contains the content moderation efforts the Foundation engaged in during the covered period. The DSA directs platforms to report meaningful and comprehensible information about the content moderation done at the platform’s own initiative.

Copyright

The Foundation publishes its Digital Millennium Copyright Act (“DMCA”) Policy to inform individuals or entities about the process by which a takedown notice can be submitted to report incidents of copyright infringement uploaded on the Wikimedia projects. Rightsholders are provided with an easy-to-understand process so they can submit their claims for evaluation.

The Foundation thoroughly evaluates each DMCA takedown request to ensure that it is valid. In this process, our Legal Department manually evaluates whether the request is consistent with the requirements of the DMCA. These requirements instruct those who claim to be copyright holders to include in their request details to enable detection of the infringement, such as the work being infringed and identification of the material believed to be infringing. We only remove allegedly infringing content when we believe that a request is valid, and we are transparent about that removal. If we do not believe a request to be valid, we will respond to the claimant providing our reasons for denying the takedown request. 

During the covered period of this report, the Foundation received 0 DMCA takedown notice(s) from the EU.

Child safety actions

While child sexual abuse material (CSAM) has been found on Wikimedia projects, it is very rare. During the covered period of this report, the Wikimedia Foundation removed 211 files as actual or suspected CSAM through our standard reporting system. The Foundation’s Trust & Safety team monitors the reports submitted through this system and also utilizes PhotoDNA, an automated tool, to identify known CSAM images and videos and report them to the National Center for Missing and Exploited Children (NCMEC), a nonprofit that refers cases to law enforcement agencies around the world. 

More on this topic can be found under Child Safety Reports in the Requests for content alteration and takedown section and the Foundation’s Combating Online Child Exploitation Policy.

Out-of-court settlements

During the covered period, 1 dispute was submitted to a DSA dispute settlement body. It was dismissed with the dispute settlement body noting that an out-of-court dispute settlement procedure is only available against an Office Action. Content moderation efforts done through volunteer community action fall outside of the DSA’s scope. The matter was resolved in 106 days with it concluding outside of this report’s covered period. Out-of-court dispute settlement procedures are detailed in Article 21 of the DSA.

Complaints through internal complaint-handling systems

Users directly involved in cases closed by the Foundation under its Office Action policy may appeal their cases. The appeals are handled by volunteers forming the Case Review Committee (CRC) in close collaboration with designated Foundation staff. The CRC may take up to 14 business days to review the appeals found to be eligible. More information about the CRC can be found here

During the relevant period, we received 10 complaints through our internal appeals system. 

Reporting on complaints through internal complaint-handling systems can be found under Article 15(d) in relation to Article 20 of the DSA.

CRC CasesNumberType
Appeals eligible for review2Bans
Appeals ineligible for review2Appeals resubmitted
Out of scope6Not within Office Action Policy

Bans and suspensions

From time to time, the Wikimedia Foundation issues Global Bans, which bar an individual from continuing to contribute to the Wikimedia projects. In the vast majority of cases, we do not know the location in which these individuals are located, and whether or not the users in question are EU persons.

During the covered period of this report, we issued bans against 34 accounts for the provision of manifestly illegal content, including harassment, and child exploitation material.

Automated content moderation

We are required to publish information about automated content moderation means, including qualitative descriptions and specified purposes of the tools, data on the accuracy of the automated tools, and descriptions of any safeguards applied, per Article 15(e) of the DSA.

The Foundation seeks to empower users to participate in content moderation processes by providing them expanded access to machine learning tools that they can use to improve or quickly remove content–these tools are developed, used, and maintained by community members. These tools primarily serve anti-vandalism, anti-spam, and anti-abuse functions. While automated tools are used to support existing community moderation processes, the bulk of the work is still done manually by human volunteers.

The tools that editors can use in the exercise of community moderation include:

  • ClueBot NG (for English Wikipedia), an automated tool used by volunteers that uses a combination of different machine learning detection methods and requires a high confidence level to automatically remove vandalism on the projects. Other similar bots include SaleBot (French, Portuguese), SeroBot (Spanish), and PatrocleBot (Romanian).
    • ClueBot NG and the other bots revert thousands of damaging edits every day from their respective language versions of Wikipedia.
    • Everything ClueBot NG does is publicly logged and easily reversible. There is an easy way for anyone to report false positives and a group of human volunteers that review all false positive reports.
  • Automoderator, is an automated anti-vandalism tool used by volunteers to reduce moderation backlog by reverting damaging edits and preventing them from entering human patroller queues. It behaves similarly to ClueBot NG and other bots, but is available to more language communities. This tool aims to give moderators confidence that automoderation is reliable and is not producing significant false positives, and to ensure that editors caught in a false positive have clear avenues to flag the error or have their edit reinstated.
    • Automoderator’s activities dashboard is available for viewing for anyone with an account. This dashboard includes the potential false positive reverts and percentage rates. To ensure that reverted editors who were making a good faith change are well equipped to understand why they were reverted, and to report false positives, Automoderator has an optional feature to send every reverted user a talk page message. 
    • Automoderator is configured by volunteers so that the community can define how cautious the model should be. Automoderator reverts edits which machine learning models determine to have a very high probability of being reverted by a volunteer editor. The criteria which Automoderator uses to determine the definition of “very high” can be customized by each language community which uses the software. To ensure that there is a standard to how the tool performs, the Foundation tested Automoderator’s accuracy to study the number of reverts it would perform depending on the caution levels defined. As a result of this testing, Automoderator’s software prevents it from being configured by communities in a way that would produce an excessive volume of false positive reverts.
  • We also create and host a series of machine learning models  which assign scores to edits and articles in order to help human editors improve articles. These models detect edits that may be damaging. Human users, ‘patrollers,’ review the flagged edits and make the final determinations about the edit quality.
  • Additionally, human users with special privileges have access to the AbuseFilter extensions, which allow them to set specific controls and create automated reactions for certain behaviors.

The Foundation Trust & Safety team also uses select automated tools such as PhotoDNA to scan for CSAM. PhotoDNA is a widely-used industry-standard tool developed by another company. Being that it is proprietary in nature, the Foundation does not possess information about its indicators of accuracy and possible rate of error. 

The volunteer community is highly effective at removing illegal and harmful content on the projects. In 2019, researchers at the Berkman Klein Center for Internet and Society at Harvard University found that the median amount of time harmful content (including harassment, threats, or defamation) remained on English language Wikipedia was 61 seconds.

Human resources

The vast majority of human resources devoted to content moderation come from the communities of independent volunteer editors, not from Wikimedia Foundation staff. Information about current active editors by language of Wikipedia, including official languages of EU member states, can be found on Wikistats.

The Wikimedia Foundation employs a staff of 10 Trust & Safety experts who are available to address complex issues requiring resolution by the Foundation. There is also the volunteer-comprised Case Review Committee that handles appeals submitted by users directly involved in cases closed by the Foundation under its Office Action policy. The Foundation does not hire outside vendors that conduct office content moderation. 

Due to safety concerns and the small size of the Trust & Safety team, we are not able to provide detailed breakdowns of their backgrounds and linguistic expertise. They are provided with quality standard trainings to ensure that the conduct of investigations and risk assessment meet a baseline of quality and non-arbitrariness. They also work together to ensure that office processes such as intake, triaging, and response are streamlined. These processes are reviewed routinely and updated as needed so that staff adapt to the demands brought on by attending to the requests and keeping up with regulatory changes.  

The team collectively provides linguistic capacity in multiple languages used on the Wikimedia projects. For EU purposes, the most relevant languages covered would be English, French, and Polish. In some cases, Trust & Safety staff may liaise with volunteer editors with competence in other languages, and/or use machine translation tools, in order to investigate and address challenges in additional languages.

Average monthly EU recipients

In order to meet the requirements of DSA Article 24(2), the Foundation created a dedicated EU DSA Userbase Statistics page to provide a reasonable estimate of monthly “active”, “unique” human users of our main project families, across the EU, averaged over a 6-month period.

Other DSA information published by the Foundation

The Foundation’s annual Audit Report, along with its Audit Implementation Report, and latest DSA Systemic Risk Assessment and Mitigation (SRAM) documentation, can be found here.