RTFP

The content moderation team is a self contained unit within Aylo, staffed entirely by employees of the company and located in Cyprus. The leader of the content moderation team reports into the Trust & Safety org of the company, which includes regulatory compliance operations, such as the DSA.

Pornhub's content moderation process includes an extensive team of human moderators dedicated to reviewing every single upload before it is published, KYC by ID verification of all third party uploaders, a thorough system for flagging, reviewing, and removing illegal material, and the utilization of a variety of automated detection technologies for known and previously identified, or potentially inappropriate content. Specifically: Hash-list tools – known illegal material We use a variety of tools that scan incoming images and videos against hash-lists provided by NGOs. If there is a match, then content is blocked before publication. • CSAI Match: YouTube’s proprietary technology for combating Child Sexual Abuse Imagery online. • PhotoDNA: Microsoft’s technology that aids in finding and removing known images of child exploitation. • Safer: In November 2020, we became the first adult content platform to partner with Thorn, allowing us to begin using its Safer product on our platforms, adding an additional layer of protection in our robust compliance and content moderation process. Safer joins the list of technologies that our platforms utilize to help protect visitors from unwanted or illegal material. • Instant Image Identifier: The Centre for Expertise on Online Sexual Child Abuse (Offlimits) tool, commissioned by the European Commission, detects known child abuse imagery using a triple verified database. • NCMEC Hash Sharing: NCMEC’s database of known CSAM hashes, including hashes submitted by individuals who fingerprinted their own underage content via NCMEC’s Take It Down service. • StopNCII.org: A global initiative (developed by Meta & SWGfL) that prevents the spread of non-consensual intimate images (NCII) online. If any adult (18+) is concerned about their intimate images (or videos) being shared online without consent, they can create a digital fingerprint of their own material and prevent it from being shared across participating platforms. • Internet Watch Foundation (IWF) Hash List: IWF’s database of known CSAM, sourced from hotline reports and the UK Home Office’s Child Abuse Image Database. AI tools – unknown illegal material We utilise several tools that use AI to estimate the ages of performers. The output from these tools assists content moderators in their decision allow publication of uploaded content. Specifically: • Google Content Safety API: Google's artificial intelligence tool that helps detect illegal imagery. • Age Estimation: We also utilize age estimation capabilities to analyze content uploaded to our platform using a combination of internal proprietary software and external technology, provided by AWS and PrivateID to strengthen the varying methods we use to prevent the upload and publication of potential or actual CSAM. Fingerprinting tools In addition to hashes received from NGOs, we also use fingerprint databases to prevent previously prohibited material from being re-uploaded. Images and videos removed during the moderation process, or subsequently removed post publication are fingerprinted using the following tools to prevent re-publication. Content may also be proactively fingerprinted with these tools. • Safeguard: Safeguard is Aylo’s proprietary image recognition technology designed with the purpose of combatting both child sexual abuse imagery and non-consensual content, by preventing the re-uploading of previously fingerprinted content to our platform. • MediaWise: Vobile’s fingerprinting software that scans any new uploads for potential matches to unauthorized materials to protect previously fingerprinted videos from being uploaded/re-uploaded to the platform. Other tools Transcription – Our proprietary audio transcription technology transcribes all content audio, which is run against our Banned Word Service. Watermarks – We utilise a proprietary watermark detection technology to help automatically identify specific terms or brands visually appearing in content.

The content moderation team maintains periodically updated capacity planning dashboards, measuring out workload and corresponding estimated resource requirements. It is staffed based on these estimates. The methodology uses actual incoming workload, estimated time needed to complete the work in units of time, and from this it estimates approximately how many human resources are required across the cumulative estimates from all moderation team daily or weekly tasks

All moderators review and assess content in a wide variety of languages and employ several tools to assess this content. All metadata is scanned against our Banned Word Service which contains a library of over 40,000 terms across more than 40 languages (Including 21 EU languages) prior to reaching moderators. Moderators then employ translation tools to evaluate the metadata to ensure that the text is compliant. Audio content is assessed by moderators who either use translation/transcription tools or who understand the spoken language in the content. In cases where the audio content cannot be understood the content is rejected as we are unable to meaningfully evaluate potential compliance issues. In all cases moderation is a collaborative task where moderators are encouraged to solicit opinions from their co-workers, senior team members, leads, and managers when reviewing content.

The majority of automated tools provide information to human moderators to aid them when making a final decision on whether or not to approve a piece of content for publication. Along with our extensive moderator training program, we also seek and highlight any patterns in moderation outcomes within particular categories, or from an individual moderator. Moderation decisions that were subsequently overturned are logged and used as part of a feedback cycle with the moderation team. Content from overturned decisions is re-reviewed to look for possible patterns with the application of particular standards or guidelines, or with particular members of the moderation team. Errors relating to an individual moderator may be followed by retraining and/or disciplinary procedures as applicable and depending upon the severity of the error and according to documented internal procedures. This constant check and review cycle ensures that any errors are handled appropriately, and training and documentation are kept up to date.

See above

See above

See above

We use a combination of automated tools, artificial intelligence, and human review to help protect our community from illegal content. While all content available on the platform is reviewed by human moderators prior to publishing, we also have additional layers of moderation which audit material on our live platform for any potential violations of our Terms of Service. The accuracy of content moderation is largely unaffected by Member State language due to our extensive use of automated tools and human moderation. Internal statistics show no significant differences between languages. Offenses are largely language independent. Automated tools are used to help inform our Trust & Safety and human moderation teams in making a manual decision. For example, when an applicable automated tool detects a match between an uploaded piece of content to one in a hash list of previously identified illegal material, an internal human verification process takes place prior to the decisioning of the particular piece of content.

We use two different virtual care platforms (North America & Europe) that give moderators access to a variety of health and wellness professionals. We also use an additional program which provides moderators with further, complementary support and tailored wellness programs consisting of fitness/nutrition/life coaches, and counsellors.

All moderators receive extensive training over a 3-month period that involves theoretical and practical exercises, job shadowing, and a final exam that requires a perfect score to pass. Once the fundamentals of the compliance guidelines are confirmed the moderators are then supervised on all their review for a period of time. Any moderation errors are addressed and corrected to ensure consistent application of the guidelines