18 Policy Brief No. 78  Toda Peace Institute Social Media companies should consider applying a tiered approach to their use of “protected characteristics”, similar to the way Facebook introduced a tiered approach to its hate speech definition and enforcement. By using different tiers, companies could maintain global policy standards, while still accounting for differences in the threat levels experienced by different “protected characteristics” across contexts and time. The designation of high risk “Protected Characteristics” could support the broadening of restriction to dangerous speech which may otherwise fall under a “grey area”, as well as trigger additional resourcing towards detection and moderation. Escalations in status should be temporary, however, and subject to a clear designation protocol and objective criteria, informed by the companies’ country-level human rights impact assessment and mitigation strategies. 4. Scale, regularly audit, and continuously improve country-level enforcement Policies need to go hand in hand with enforcement if they are to make a difference. Since 2018, Facebook has worked hard to scale its enforcement capacity in Myanmar, hiring additional content reviewers as well as rolling out AI classifiers for Burmese language. Though this has made the company more proactive in its moderation, a number of major challenges remain. Independent research by the author, for example, identified various flaws with Facebook’s report triage system, which accounts for systematic and widespread content moderation errors on user reports. This is a serious issue as user reports remain the only path to content moderation for Myanmar’s minority users – as well as for a significant number of users globally, whose languages are yet to leverage AI classifiers. Though AI classifiers are now contributing to the detection of potentially violating content in Burmese language, the accuracy of these classifiers also continues to be widely inadequate, resulting in undue sanctions being regularly applied, at times with significant implications for freedom of expression and information.  Social Media companies should conduct regular market level audits of their content moderation and enforcement processes. Particular attention should be given to reporting flows and triaging systems, AI classifiers, and the recruitment and activity of human reviewers. Where possible, these audits should be conducted by an independent third party, possibly overseen by an independent oversight body, such as the newly established Facebook Oversight Board.  Social Media companies should also contribute funding and data to support the development of ethnic language corpora and consider developing scholarships in Natural Language Processing (NLP) and AI targeted at minority and ethnic language communities. 5. Scale efforts to detect and curtail abuse at its source Though increased capacity for content level moderation is important, it is not enough to deter and root out coordinated abuse. Bad actors spend months—if not years—building networks of online assets, including accounts, pages and groups, that allow them to manipulate the conversation. These inauthentic presences continue to present a major risk in places like Myanmar and are responsible for the overwhelming majority of problematic content.