A/HRC/38/35 Rule enforcement 56. Automation and human evaluation. Automated content moderation, a function of the massive scale and scope of user-generated content, poses distinct risks of content actions that are inconsistent with human rights law. Company responsibilities to prevent and mitigate human rights impacts should take into account the significant limitations of automation, such as difficulties with addressing context, widespread variation of language cues and meaning and linguistic and cultural particularities. Automation derived from understandings developed within the home country of the company risks serious discrimination across global user bases. At a minimum, technology developed to deal with considerations of scale should be rigorously audited and developed with broad user and civil society input. 57. The responsibility to foster accurate and context-sensitive content moderation practices that respect freedom of expression also requires companies to strengthen and ensure professionalization of their human evaluation of flagged content. This strengthening should involve protections for human moderators consistent with human rights norms applicable to labour rights and a serious commitment to involve cultural, linguistic and other forms of expertise in every market where they operate. Company leadership and policy teams should also diversify to enable the application of local or subject-matter expertise to content issues. 58. Notice and appeal. Users and civil society experts commonly express concern about the limited information available to those subject to content removal or account suspension or deactivation, or those reporting abuse such as misogynistic harassment and doxing. The lack of information creates an environment of secretive norms, inconsistent with the standards of clarity, specificity and predictability. This interferes with the individual’s ability to challenge content actions or follow up on content-related complaints; in practice, however, the lack of robust appeal mechanisms for content removals favours users who flag over those who post. Some may argue that it will be time-consuming and costly to allow appeals on every content action. But companies could work with one another and civil society to explore scalable solutions such as company-specific or industry-wide ombudsman programmes. Among the best ideas for such programmes is an independent “social media council”, modelled on the press councils that enable industry-wide complaint mechanisms and the promotion of remedies for violations. 140 This mechanism could hear complaints from individual users that meet certain criteria and gather public feedback on recurrent content moderation problems such as overcensorship related to a particular subject area. States should be supportive of scalable appeal mechanisms that operate consistently with human rights standards. 59. Remedy. The Guiding Principles highlight the responsibility to remedy “adverse impacts” (principle 22). However, few if any of the companies provide for remediation. Companies should institute robust remediation programmes, which may range from reinstatement and acknowledgment to settlements related to reputational or other harms. There has been some convergence among several companies in their content rules, giving rise to the possibility of inter-company cooperation to provide remedies through a social media council, other ombudsman programmes or third-party adjudication. If the failure to remediate persists, legislative and judicial intervention may be required. 60. User autonomy. Companies have developed tools enabling users to shape their own online environments. This includes muting and blocking of other users or specific kinds of content. Similarly, platforms often permit users to create closed or private groups, moderated by users themselves. While content rules in closed groups should be consistent with baseline human rights standards, platforms should encourage such affinity-based groups given their value in protecting opinion, expanding space for vulnerable communities and allowing the testing of controversial or unpopular ideas. Real-name requirements 140 18 See ARTICLE 19, Self-regulation and ‘Hate Speech’ on Social Media Platforms (London, 2018), pp. 20−22.