Failures in Moderating Brazilian Pro-Coup Content

The final speakers in this session at AoIR 2023 are Marcel Alves dos Santos Jr. and, again, Emilie de Keulenaar (and I’m on 2% charge, so let’s see how far we get here). Marcel begins by pointing to Brazil’s unresolved relationship with its past military dictatorships: its Constitution of 1988 was accompanied by an amnesty for members of the military who were implicated in human rights abuses.

These issues were brought to the forefront again during the imprisonment of former president Lula da Silva and the presidency of former soldier Jair Bolsonaro, which emboldened military leaders to involve themselves in democratic matters, yet since 2021 there is also a law against the glorification of military rule. The present project, therefore, examines how Twitter, Meta, and YouTube moderated content relating to Brazilian militarism during the 2022 Brazilian elections, and how such moderation approaches align with Brazil’s own approaches to moderating militaristic rhetoric (when platform rules usually merely translate US-centric moderation rules to other country contexts).

The project engaged in the dynamic archiving of data from a number of different platforms, including some 450 far-right Brazilian channels on YouTube as well as collections of militaristic rhetoric across various other platforms, to identify deleted content and determine the reason given for its deletion. This can also trace the rhetoric of such content over time, and the engagement it received on various platforms. It shows the gradual building of far-right pressure for a coup, which eventuated in January 2022. Coup plotting spread from Telegram to more mainstream platforms, and mostly received no moderation other than some labelling or (after the failure of the coup) self-moderation by people involved in the coup attempt.

On YouTube, some 20% of the far-right videos were deleted during the election period itself; videos posted after the election – which engaged in calling for and planning of the coup – were more likely to remain online, which might indicate that platforms no longer paid as much attention to moderation after the election concluded, or moderated for the wrong keywords. Only after the 8 January coup did moderation kick in again more prominently. Channels that live-streamed the attack on Brasilia, in particular, now no longer exist.

Meta’s own Oversight Board has recognised the platform’s failure to moderate content calling for a coup, in fact: focused on misinformation, its systems failed to identify the problematic nature of the coup calls. Research into takedowns is critical in retracing such failures. (Yay, made it – even if I had to write the final paragraphs on my phone.)