Opening Remarks

X was founded on a commitment to transparency. We also want people on X to feel they are able to freely express themselves, while also ensuring that conversations on X are safe, legal and unregretted. When you think about some of the world’s most powerful moments, movements, and memes, they prevailed because people had a place to express their ideas, challenge conventional norms, and demand better. That’s why free expression matters.

We also believe, and we’re proving, that free expression and platform safety can coexist. X is reflective of real conversations happening in the world, and that sometimes includes perspectives that may be offensive, controversial, and/or narrow-minded to others. While we welcome everyone to express themselves on X, we will not tolerate behaviour that harasses, threatens, dehumanises or uses fear to silence the voices of others. Our TIUC Terms of Service and Rules - which are continually being reviewed, and are informed by feedback from the people who use X - help ensure everyone feels safe expressing themselves.

We are committed to fair, informative, responsive, and accountable enforcement. In the past, we too often got caught in a binary paradigm of whether to leave content up, or take it down.

• The risks of getting it wrong at the extremes are great: on one hand, you can leave up content that’s really dangerous; on the other, you run the risk of censorship.
• Our point is: if you do either, you need to be right. And we live in a world with many shades of grey.

To be clear, we do continue to remove dangerous and illegal content and accounts. X also responds to reports of illegal content and takes action on content that violates local laws. But what we’ve learned is that there are other types of content where a range of potential reasonable, proportionate, and effective approaches, that also seek to balance fundamental rights, can be appropriate.

You can think about how we moderate on X in three buckets: content and accounts that remain, are restricted, and are removed.

1. Remain: The overwhelming majority of content on X is healthy—meaning it does not violate our TIUC Terms of Service and Rules or our policies such as Hateful Conduct, Abuse & Harassment, and more. Keep in mind: just because a post doesn’t violate a policy, doesn’t mean everyone will like it.
2. Restrict: This is where our new Freedom of Speech, Not Reach enforcement philosophy is used. For content that may be interpreted as potentially violating our policies—meaning it’s awful, but lawful—we restrict the reach of posts by making the content less discoverable, and we’re making this action more transparent to everyone. When we decide to restrict a piece of content, a restricted reach label is applied, the ability to engage with the content is taken away, and its reach is restricted to views occurring directly on the author's profile. Restricted reach labels are not in use for all policies; our restricted reach labels were initially only applied to Hateful Conduct, but we have since expanded application to our Abuse & Harassment, Civic Integrity, and Violent Speech Policies. That said, restricting content—or even a whole account—is something we’ve done for a long time, and we have a range of enforcement options for the variety of use cases that we face every day. For example, we may also place an account in read-only mode, temporarily limiting its ability to post, Repost, or Like.
3. Remove: If reported content is illegal, we withhold access to it in the respective jurisdictions. We also know that certain types of content, such as targeted violent threats, targeted harassment, or privacy violations, can be extremely harmful if not removed and we either suspend outright or require that this content be deleted before returning to the platform.

We've made significant progress towards improving the safeguards to protect our users and our platform, but we know that this critical work will never be done. X is committed to ensuring the safety and health of the platform and fulfilment of its DSA Compliance obligations through our continued investment in human and automated protections.

This report covers the content moderation activities of X’s international entity Twitter International Unlimited Company (TIUC) under the Digital Services Act (DSA), during the date range August 28, 2023 to October 20, 2023.

We refer to “notices” as defined in the DSA as “user reports” and “reports”.

Description of our Content Moderation Practices

X's purpose is to serve the public conversation. Violence, harassment, and other similar types of behaviour discourage people from expressing themselves, and ultimately diminish the value of global public conversation. Our rules are designed to ensure all people can participate in the public conversation freely and safely.

X has policies protecting user safety as well as platform and account integrity. The X Rules and Policies are publicly accessible on our Help Center, and we are making sure that they are written in an easily understandable way. We also keep our Help Center regularly updated anytime we modify our rules.

Additionally, you will find explanations in our Help Center on our policy development process and rules enforcement philosophy. Creating a new policy or making a policy change requires in-depth research around trends in online behaviour, developing clear external language that sets expectations around what’s allowed, and creating enforcement guidance for reviewers that can be scaled across millions of pieces of content and accounts. Our policies are dynamic, and we continually review them to ensure that they are up-to-date, necessary, and proportional.

We consider diverse perspectives around the changing nature of online speech, including how our Rules are applied and interpreted in different cultural and social contexts. We then test the proposed rule with samples of potentially violative content to measure the policy effectiveness, and once we determine it meets our expectations, we build and operationalise product changes to support the update. Finally, we train our global review teams, update the X Rules, and start enforcing the relevant policy.

While we aim to enable open discussion of differing opinions and viewpoints, we are committed to the objective, timely, and consistent enforcement of our rules. This approach allows many forms of speech to exist on our platform and, in particular, promotes counterspeech: speech that presents facts to correct misstatements or misperceptions, points out hypocrisy or contradictions, warns of offline or online consequences, denounces hateful or dangerous speech, or helps change minds and disarm.

Thus, context matters. When determining whether to take enforcement action, we may consider a number of factors, including (but not limited to) whether:

• The behaviour is directed at an individual, group, or protected category of people;
• The report has been filed by the target of the abuse or a bystander;
• The user has a history of violating our policies;
• The severity of the violation;
• The content may be a topic of legitimate public interest.

When we take enforcement actions, we may do so either on a specific piece of content (e.g., an individual post or Direct Message) or on an account. We may employ a combination of these options. In most cases, this is because the behaviour violates the X Rules.

X strives to provide an environment where people can feel free to express themselves. If abusive behaviour happens, we want to make it easy for people to report it to us. EU users can also report any violation of our rules or their local laws, no matter where such violations appear, and we’ve recently improved our reporting flow to make it easier to use in several key ways. It now takes less steps to report most content, with extra steps only when it helps us take the right action. We now have clearer choices that match directly to our policies and how they’re communicated externally. We’ve also included new options that were previously only available at help.x.com.

EXERCISE OF MODERATION

To enforce our rules, we are using a combination of machine learning and human review. Our systems are able to surface content to human moderators who use important context to make decisions about potential rule violations. This work is led by an international, cross-functional team with 24-hour coverage and the ability to cover multiple languages. We also have a complaints process for any potential errors that may occur.

Examples of actions we may take:

• Placing a post behind a notice: We may place some forms of sensitive media like adult content or graphic violence behind an interstitial advising viewers to be aware that they will see sensitive media if they click through. We also give users the option to control whether they see sensitive media.
• Withholding a post based on age: We restrict views of specific forms of sensitive media such as adult content for viewers who are under 18, or who do not include a birth date on their profile, with interstitials. 
• Withholding a post or account in a country: We may withhold access to certain content in a particular country if we receive a valid and properly scoped request from an authorised entity in that country. Read more about country withheld content.

To ensure that our human reviewers are prepared to perform their duties we provide them with a robust support system. Each human reviewer goes through extensive training and refreshers, they are provided with a suite of tools that enable them to do their jobs effectively, and they have a suite of wellness initiatives available to them. For further information on our human review resources, see the section titled “Human resources dedicated to Content Moderation”.

We always aim to exercise moderation with transparency. Where our systems or teams take action against content or an account as a result of violating our rules or in response to a valid and properly scoped request from an authorised entity in a given country, we strive to provide context to users. Our Help Center article explains notices that users may encounter following actions taken. We will also promptly notify affected users about legal requests to withhold content, including a copy of the original request, unless we are legally prohibited from doing so.

COOPERATION WITH PUBLIC AUTHORITIES

Cooperation with law enforcement authorities within the EU is crucial to X. We work closely with law enforcement, and we do our best to assist them in identifying users whose content may be in violation of local laws. Any law enforcement authority or agency can find guidelines on our Help Center specifically for law enforcement and can reach out to X using a dedicated form.

TIUC is headquartered in Dublin, Ireland, and processes law enforcement requests relating to users who live in the EU. We receive and respond to requests related to user data from EU law enforcement agencies and judicial authorities wherever there is a valid legal process. We have existing processes in place, including a dedicated online portal for law enforcement, and expert teams with global coverage across all timezones that review and respond to reports in diverse languages.

Law enforcement can use our dedicated portal to submit their legal demands and can request the following information:

• Information requests: Requests/orders for user personal and private information (e.g Basic Subscriber Information).
• Content removal requests: Requests/orders to remove content based on TIUC Terms of Service and Rules or EU local laws.
• Preservation requests: Requests to preserve data for the purposes of an investigation.
• Emergency requests: Process through which, when there is an imminent threat to life or serious bodily harm, X may disclose user information to law enforcement without receiving formal legal process.

Our Own Initiative Content Moderation Activities

AUTOMATED CONTENT MODERATION

X employs a combination of heuristics and machine learning algorithms to automatically detect content that violates the X Rules and policies enforced on our platform.

MACHINE LEARNING MODELS

We use combinations of natural language processing models, image processing models and other sophisticated machine learning methods to detect potentially violative content. These models vary in complexity and in the outputs they produce. For example, the model used to detect abuse on the platform is trained on abuse violations detected in the past. Content flagged by these machine learning models are either reviewed by human content reviewers before an action is taken or, in some cases, automatically actioned based on model output.

HEURISTIC MODELS

Heuristics are typically utilised to enable X to react quickly to new forms of violations that emerge on the platform. Heuristics are common patterns of text or keywords that may be typical of a certain category of violations. Pieces of content detected by heuristics may also get reviewed by human content reviewers before an action is taken on the content. These heuristics are used to flag content for review by human agents and prioritise the order such content is reviewed.

TESTING, EVALUATION, AND ITERATION

Automated enforcements under the X Rules and policies undergo rigorous testing before being applied to the live product. Both machine learning and heuristic models are trained and/or validated on thousands of data points and labels (e.g., violative or non-violative) that are generated by trained human content reviewers. For example, inputs to content-related models can include the text within the post itself, the images attached to the post, and other characteristics. Training data for the models comes from both the cases reviewed by our content moderators, random samples, and various other samples of pieces of content from the platform.

Once reviewers have confirmed that the detection meets an acceptable standard of accuracy, we consider the automation to be ready for launch. Once launched, automations are monitored dynamically for ongoing performance and health. If we detect anomalies in performance (for instance, significant spikes or dips against the volume we established during sizing, or significant changes in user complaint/overturn rates), our Engineering (including Data Science) and Policy teams revisit the automation to diagnose any potential problems and adjust the automations as appropriate.

USE OF HUMAN MODERATION

Before any given algorithm is launched to the platform, we verify its detection of policy violating content or behaviour by drawing a statistically significant test sample and performing item-by-item human review. Reviewers have expertise in the applicable policies and are trained by our Policy teams to ensure the reliability of their decisions. During this testing phase, we also calculate the expected volume of moderation actions a given automation is likely to perform in order to set a baseline against which we can monitor for anomalies in the future (called “sizing”). Human review helps us to confirm that these automations achieve a level of precision, and sizing helps us understand what to expect once the automations are launched.

In addition, humans proactively conduct manual content reviews for potential policy violations. We conduct proactive sweeps for certain high-priority categories of potentially violative content both periodically and during major events, such as elections. Agents also proactively review content flagged by heuristic and machine learning models for potential violations of other policies, including our sensitive media, child sexual exploitation (CSE) and violent and hateful entities policies.

AUTOMATED MODERATION ACTIVITY EXAMPLES

A vast majority of all accounts that are suspended for the promotion of terrorism and CSE are proactively flagged by a combination of technology and other purpose-built internal proprietary tools.

When we remove CSE content, we immediately report it to the National Center for Missing and Exploited Children (NCMEC). NCMEC makes reports available to the appropriate law enforcement agencies around the world to facilitate investigations and prosecutions.

Our current methods for surfacing potentially violative terrorist content for review include leveraging the shared industry hash database, e.g., supported by the Global Internet Forum to Counter Terrorism (GIFCT), and deploying a range of internal tools and/or utilising the industry hash sharing (e.g., PhotoDNA) prior to any reports filed. We commit to continuing to invest in technology that improves our capability to detect and remove, for instance, terrorist and violent extremist content online, including the extension or development of digital fingerprinting and AI-based technology solutions. Our participation in multi-stakeholder communities, such as the Christchurch Call to Action, Global Internet Forum to Counter Terrorism and EU Internet Forum (EUIF), helps to identify emerging trends in how terrorists and violent extremists are using the Internet to promote their content and exploit online platforms.

You can learn more about our commitment to eradicating CSE and terrorist content, and the actions we’ve taken here. Our continued investment in proprietary technology is steadily reducing the burden on people to report this content to us.

SCALED INVESTIGATIONS

These moderation activities are supplemented by scaled human investigations into the tactics, techniques and procedures that bad actors use to circumvent our rules and policies. These investigations may leverage signals and behaviours identifiable on our platform, as well as off-platform information, to identify large-scale and/or technically sophisticated evasions of our detection and enforcement activities. For example, through these investigations, we are able to detect coordinated activity intended to manipulate our platform and artificially amplify the reach of certain accounts or their content.  

CLOSING STATEMENT ON CONTENT MODERATION ACTIVITIES

Our content moderation systems are designed and tailored to mitigate systematic risks without unnecessarily restricting the use of our service and fundamental rights, especially freedom of expression. Content moderation activities are implemented and anchored on principled policies and leverage a diverse set of interventions to ensure that our actions are reasonable, proportionate and effective. Our content moderation systems blend automated and human review paired with a robust appeals system that enables our users to quickly raise potential moderation anomalies or mistakes.

Enforcement Activity Summary Data

RESTRICTED REACH LABELS DATA: FREEDOM OF SPEECH, NOT REACH

Our mission at X is to promote and protect the public conversation. We believe X users have the right to express their opinion and ideas without fear of censorship. We also believe it is our responsibility to keep users on our platform safe from content that violates our rules.

These beliefs are the foundation of Freedom of Speech, Not Reach - our freedom of expression based enforcement philosophy which means, where appropriate, restricting the reach of posts that are classified as potentially meeting our threshold for enforcement under our Hateful Conduct, Abuse & Harassment, Civic Integrity, and Violent Speech policies. Please note these policies have a range of enforcement actions, such as removal, suspension, and restricted reach.

Restricting the reach of posts, also known as visibility filtering, is one of our existing enforcement actions that allows us to move beyond the binary “leave up versus take down” approach to content moderation. Posts with these labels will be made less discoverable on the platform. This can include:

• Excluding the post from search results, trends, and recommended notifications;
• Removing the post from the For you and Following timelines;
• Restricting the post’s discoverability to the author’s profile;
• Downranking the post in replies;
• Restricting Likes, replies, Reposts, Quote posts, bookmarks, share, pin to profile, or Edit post.

Additionally, these labels bring transparency to this enforcement action by displaying which policy the post potentially violates to both the author and other users on X, and communicating that the post’s visibility is limited. Authors can submit a complaint on the label if they think we incorrectly limited their post’s visibility.

RESTRICTED REACH LABELS DATA

Important Note: The table lists actions of visibility filtering on content potentially violative of our rules in accordance with our Freedom of Speech, Not Reach enforcement philosophy. We did not apply any visibility filtering based on illegal content.

ACTIONS TAKEN ON CONTENT FOR TIUC TERMS OF SERVICE AND RULES VIOLATIONS

ACTIONS TAKEN ON ACCOUNTS FOR TIUC TERMS OF SERVICE AND RULES VIOLATIONS

Important Notes about Action based on TIUC Terms of Service and Rules Violations:

1. The categories “Other” refer to cases of workflow exceptions and tooling inconsistencies which prevent a further clarification on the violated policy of TIUC Terms of Service and Rules.
2. User reports of illegal content which have been actioned under TIUC Terms of Service and Rules are displayed in the table "Actions Taken on Illegal Content".

*A data extraction limitation is impacting the availability of data ranging from Aug 28 to Sept 23. See the table "TIUC Terms of Service and Rules Content Removal Actions - Sep 5 to Sep 23” in the Appendix.

Orders received from Member States’ authorities including orders issued in accordance with Articles 9 (Removal Orders) and 10 (Information Requests)

REMOVAL ORDERS, Art. 9 DSA

Important Notes about Removal Orders:

1. To improve clarity, we've omitted countries and violation types with no legal requests from the tables above.
2. The table “Removal Orders Median Handle Time” shows the category which we considered to fit best and under which we handled the order. This category might deviate from the information provided by the authority when submitting the order via the X online submission platform.
3. In the cases from France and Spain, we asked the submitting authority to fulfil Article 9 information requirements but did not receive responses in the reporting period.

INFORMATION REQUESTS, Art. 10 DSA