Written evidence submitted by X (SMH0064)
X submission to Commons Science, Innovation and Technology Committee Inquiry: “social media, misinformation and the role of algorithms”
Dear Chair and Members for the Science, Innovation and Technology Committee,
Thank you for the opportunity to provide written evidence to the Commons Science, Innovation and Technology Select Committee Inquiry: Social media, misinformation and the role of algorithms.
At X, our purpose is to serve the public conversation, ensuring a safe environment where everyone can participate freely and confidently. X’s approach to Safety is underpinned by our Terms of Service, including the X Rules and Policies, which are designed to ensure all of our users can participate in the public conversation freely and safely.
X is reflective of real conversations happening in the world and that sometimes includes perspectives that may potentially be offensive or controversial to others. Our platform empowers people to understand different sides of an issue and therefore encourages dissenting opinions and viewpoints to be discussed openly. 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.
While we welcome everyone to express themselves on our platform, we do not tolerate behavior that harasses, threatens, or uses fear to silence the voices of others - such behaviour discourages people from expressing themselves, and ultimately diminishes the value of the global public conversation. We are committed to maintaining an inclusive environment, where diverse perspectives can be shared, debated and appreciated by our users.
X has a dedicated portal for law enforcement. This includes both standard information requests from law enforcement as well as emergency requests which are handled with urgency, in accordance with legal requirements. Both of these avenues are open for any requests from law enforcement agencies. Guidelines for law enforcement authorities are readily available on our Help Centre.
X remains fully committed to compliance with the duties set out in the Online Safety Act and we welcome our continued engagement with Ofcom. Our approach to safety, as set out in this submission, demonstrates X’s own initiative and independent commitment protecting users. A key priority for X is working towards Online Safety Act regulatory deliverables, in line with Ofcom’s roadmap. Our reporting features mentioned above provide clear policies designed to prevent harmful content and behaviours on our platform. We are committed to transparency and accountability, regularly reviewing and reporting our progress to Ofcom and stakeholders. We aim to enhance user trust and ensure a positive experience for all of our users.
X’s approach to content moderation
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 by reference to principled policies and leverage a diverse set of interventions to ensure that our actions are reasonable, proportionate and effective.
We have been taking an extensive and holistic approach towards freedom of expression by investing in developing a broader range of remediations beyond the binary “leave up versus take down” approach to content moderation, with a particular focus on education, rehabilitation, and deterrence. With this in mind, we moderate content on X in three buckets:
Restricted reach labels are applied to content that may be in violation of our Hateful Conduct, Abuse & Harassment, Civic Integrity, and Violent Speech Policies. We may also place an account in read-only mode, temporarily limiting its ability to post, repost, or Like.
To enforce our Rules, we use 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 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. Our complaints process is designed to be user-friendly and efficient, allowing for the prompt resolution of issues and fostering trust within our community. We strive to uphold the highest standards of safety and user protection on our platform.
X is committed to transparency and our latest Global Transparency Report covers X Safety Enforcement for the period of January to June 2024.
Authenticity
On X, our policies require users not to share inauthentic content that may deceive others or lead to harm. This includes the sharing of inauthentic media, including, manipulated, or out-of-context media that may result in widespread confusion on public issues, impact public safety, or cause serious harm (“misleading media”).
We may use our own technology or receive reports through partnerships with third parties in order to determine if media have been manipulated or presented out of context. Content that we determine as misleading or out-of-context media will be given a publicly visible label that is transparent to all users. In situations where we are unable to reliably determine that media is misleading, we may not take any action.
Our Synthetic And Manipulated Media policy prohibits users from sharing media that is shared as authentic but is significantly and deceptively altered, manipulated, or fabricated in a way that fundamentally changes its meaning and can result in widespread confusion on public issues, impact public safety, or cause serious harm, such as:
● Manipulated: Media that have been substantially edited or post-processed in a manner that fundamentally alters their composition, sequence, timing, or framing and distorts their meaning; media with visual or auditory information (such as new video frames, overdubbed audio, or modified subtitles) that has been added, edited, or removed that fundamentally changes the understanding, meaning, or context of the media; media created, edited, or post-processed with enhancements or use of filters that fundamentally changes the understanding, meaning, or context of the content; and media depicting a real person that has been fabricated or simulated, especially through use of algorithms or broader artificial intelligence.
● Out-of-context: Media that is not manipulated, but is shared in a deceptive manner or out-of-context or with intent to deceive people about the nature or origin of the content and can result in widespread confusion on public issues, impact public safety, or cause serious harm, such as: media presented with false or misleading context surrounding the source, location, time, or authenticity of the media; media presented with false or misleading context surrounding the identity of the individuals or entities visually depicted in the media; and media presented with misstatements or misquotations of what is being said or presented with fabricated claims of fact of what is being depicted.
Please note, X allows the sharing of manipulated or out-of-context media in non-deceptive ways.
X has long had policies designed to safeguard the platform from bad actors, including teams that address coordinated accounts and similar problems. That remains true - and in addition, we believe it’s important to add context to potentially misleading content.
Further details on X’s approach to authenticity are available here.
Community Notes
X's Community Notes product aims to create a better informed world by empowering people on X to collaboratively add context to potentially misleading posts. Contributors can leave notes on any post and if enough contributors from different points of view rate that note as helpful, the note will be publicly shown on a post.
Community Notes was created to deliver context and information in a way that people would trust and feel is fair. In order to do this, Community Notes has two unique features. First, it lets a diverse group of contributors on X propose context to add to posts, and shows notes broadly across X when they are found helpful by people who’ve disagreed with each other in the past. As a result, notes cover a wide range of topics, and are found helpful across the political spectrum, even on contentious topics. Second, Community Notes’ code and data are open-source, allowing anyone to view its logic to analyse, critique, or help improve it. This makes it an extraordinarily transparent and unique approach to addressing misleading information.
There are now over 900,000 contributors in 197 countries to add helpful context to posts on X, including ads. Community Notes continue to play a particularly important role when it comes to highly engaged content, including in key moments and news events. We are proud to see other platforms following in our footsteps. Earlier this week, Meta announced they will adopt a Community Notes-style fact-checking model. This announcement follows YouTube’s late-2024 commitment to adopt a Community Notes model, underscoring the impact X has had on redefining industry standards.
A recent study found that, across the political spectrum, Community Notes were perceived as significantly more trustworthy than traditional, simple misinformation flags. Importantly, this study also found that Community Notes had a greater effect on improving people’s identification of misleading posts. Further independent studies show posts with notes are shared 50-61% less, and deleted 80% more.
X has also configured Community Notes to be automatically shown on posts that feature AI-generated images and other out-of-context media. The approximately 7,800 media notes that have been written are now showing on over 600,000 distinct posts and have been seen over 2.5 billion times. Anyone can now request a Community Note, and with enough requests, top contributors will be alerted and can propose notes.
We know there are many challenges involved in building an open, participatory system like Community Notes — from making it resistant to manipulation attempts, to ensuring it isn’t dominated by a simple majority or biased because of the distribution of contributors. We have intentionally designed Community Notes to mitigate potential risks, through a range of built-in safeguards such as a contributors’ vetting process and Community Notes programme specific rules to ensure that, while Community Notes is an open and participatory tool, it’s not subject to abuse.
Incident response protocols
X has internal incident response protocols in place when a high-visibility event occurs and virality triggers rapid and widespread proliferation of various content types on the platform. X has a dedicated team monitoring the platform conversation and on the ground developments. This team proactively reviews narratives on the platform against various policies, including Violent Speech, Hateful Conduct, and Synthetic and Manipulated Media.
When violative trends are found, this team works with relevant stakeholes to update platform enforcement guidelines, informs operational teams, and proactively escalates violative posts for swift actions.
For example, in response to the brutal attack that took place in Southport on Monday 29th July and the subsequent incidents that occurred across the UK, X’s Safety team were vigilant and steadfast in their commitment to keeping the platform safe.
This included, for example, active monitoring of the platform to surface content in violation of the X Terms of Service. As a result, proactive action has been taken on tens of thousands of pieces of content under our rules.
Actions included:
- Account suspension
- Content removal
- Actioning content for violations of the X Hateful Conduct and Violent Speech policies
- Labelling content as graphic, violent or sensitive media
- Labelling content as manipulated media
- Labelling content as out of context
Furthermore, Community Notes, as described above, which allows people to collaboratively add context to potentially misleading posts on X, played an important role in addressing misleading content across the platform in relation to the Southport incident and subsequent violence. Notes in relation to these incidents have been viewed millions of times.
In response to the incidents occurring across the UK following the Southport attack, we engaged with a range of external stakeholders, including meeting regularly with the National Security Online Information Team in the Department for Science, Innovation and Technology (DSIT). Additionally, we met with the Secretary of State for Technology and Ofcom to provide an update on our incident response.
To support law enforcement officers handling the response to UK civil unrest during July and August, we were pleased to provide a training session on best practices in the use of our law enforcement portal.
Recommending responsibly on X
Firstly, it is important to note that on X we give users a clear and concise choice over the recommendations they see/receive on X. People have two options to view posts: “For You” and “Following”. Under the Following tab, you only see posts from accounts you follow. Under the For You tab, you will see recommended posts for you from both your network (accounts, lists and topics you follow) as well as from accounts that you don’t explicitly follow.
Every day, people come to X to keep up with what’s happening in the world and the For You tab aims to deliver the user the best of what’s happening in the world right now. This requires a recommendation algorithm to distill the millions of daily posts on X down to a handful of top posts that ultimately show up in the For You timeline. The choices people make when using X, such as the accounts and topics they follow, help determine the content they see. For example, when a user follows an account or topic, those posts will appear in such user’s For You timeline alongside others we think the user might be interested in.
In addition to showing content from accounts and topics, we also make recommendations to make it easier and faster for users to find content and accounts that are relevant to their interests.
X recommendations are based upon a variety of signals, including, but not limited to, interests chosen during onboarding, accounts and topics followed, posts liked, reposted, or otherwise engaged with, and content that is popular in a user’s network.
Recommendations may amplify content, so it’s important they’re surfaced responsibly. Our recommender systems are designed to exclude harmful and violating content by integrating with Visibility Filtering systems on X and others. We have several ways of preventing potentially harmful or offensive content and accounts from being amplified, including using machine learning technology, reviewing user reports, and other tools.
To increase transparency and accountability, X opened sourced its algorithm on GitHub in March 2023, with some updates since that release and X’s engineering team published a Blogpost to explain to the public how X’s algorithm works.
We welcome feedback on recommendations people using X receive. For ‘For You’ timeline recommendations, feedback can be provided by selecting “Not interested in this post” or “Not interested in this Topic”. We use this as a signal to recommend less of that type of content to the user.
Additionally, controls are important in both helping people on X curate their own experience and provide critical feedback for our recommendation systems. A variety of options are available for people using X to control what they do and don’t see on our service. Features include but are not limited to, mute and block functions; the option to filter notifications; extensive privacy and safety settings; and the ability to turn off autoplay for videos.
Brand safety
Finally, we take this opportunity to touch on X’s Brand Safety policies, as well as the controls we offer people and advertisers. Illegal content and content in violation of the X Rules are not in X’s commercial interests. Our Brand Safety policies build upon the foundation laid by our Rules to promote a safe advertising experience for customers and brands. Specifically, our policies and controls are designed to avoid placement immediately above or below content that we determine may be unsafe or unsuitable, including content that falls below industry standards. Our Brand Safety policies inform the context in which we serve ads, and include, but are not limited to:
● Adult sexual content
● Hate or extremist content
● Profanity and offensive language
● Restricted and illegal products and services
● Sensitive content
● Violent, objectionable, or graphic content
For further details on X’s Brand Safety policies and how these are enforced across the platform, please see our Brand Safety Policy and its application in the Amplify Pre-Roll program.
8 January 2025