A Decade After the Rohingya Atrocities, Can Social Media Platforms Still Be Held Accountable?

A project reflection by Human Rights Myanmar, supported by the Pulitzer Center through the South to South (S2S) AI Accountability CoLab microgrant.

A decade ago, social media platforms contributed to mass atrocities against the Rohingya in Myanmar. Facebook in particular had become a major source of information, but its Ireland-based human moderators struggled with the language, context, and political realities of the country. Facebook apologized at the time.

Today, AI can process content at a far greater scale. But the core human rights question remains whether these systems protect people in Myanmar.

Our Pulitzer Center-supported project began with that question, situating Myanmar within a broader global debate about AI accountability, platform responsibility, and the protection of vulnerable communities. We wanted to know whether, a decade later and after some significant moderation U-turns following the election of President Trump in the U.S., social media platforms were once again failing to moderate hate speech.

We monitored Facebook, Telegram, and TikTok for a year and identified more than 600 posts, including 232 on Facebook, containing what we assessed as actionable hate speech. We wanted to see whether those posts would be removed naturally through the platforms’ own automated and user-generated moderation systems. For the ones that were not, we stepped in after a month, reported them, and watched what happened, building a detailed picture.

The result was clear. More than half the posts were deleted while we watched them. But when our anonymized users reported the rest, most of our reports and appeals were rejected. That does not prove that the systems are failing overall. But it does show that harmful content can remain online even after users challenge it, and that those challenges often have limited effect. Our new report, which focuses on Facebook, gives stakeholders, including Facebook itself, a clear, detailed record full of emerging patterns that are harder to dismiss than a single claim or isolated example.

The deeper problem on Facebook and other platforms is accountability. Platforms make enormous claims about the sophistication of their AI systems, yet civil society has very little information to test those claims. Facebook’s transparency data is vague and global, which tells us little about how its systems work in Myanmar or how they allow affected stakeholders to challenge the company on specific facts.

We also tried to investigate whether platforms were not just failing to remove hate speech, but actively helping spread it through “amplification.” We found many possible examples, but the evidence was not strong enough to publish, particularly given how litigious at least one platform has become towards its critics, including people condemning its Myanmar policies. It does, however, underline how little information platforms release that could hold them accountable.

There is another reason this matters. Platform policies, AI, and moderation systems are designed and governed outside Myanmar. We have repeatedly raised concerns that American platforms change dramatically in Myanmar every time a new U.S. government is elected, including just last year following the election of President Trump.

Platforms may be private companies, but there is a serious question about whether people in Myanmar should bear the human rights consequences of decisions made elsewhere without meaningful ways to understand, challenge, or influence them. The U.S. is not unaware of this issue, given the importance of “no taxation without representation” in its own colonial history.

These questions about accountability and power are not unique to Myanmar, but are central to the wider AI accountability landscape, especially where powerful systems are designed and governed far from the people they affect. The Pulitzer Center grant has enabled us to start investigating this challenge. Looking ahead, we want to build on our Facebook report and complete our analysis of the 400 hate speech posts we monitored on Telegram and TikTok. We also need stronger civil society partnerships to measure more reliably how recommendation systems “amplify” harmful content.

The goal is not to prove that AI is bad or that platforms should remove more content. It is more fundamentally to show that powerful technology companies should carry out adequate human rights due diligence. A decade after the Rohingya atrocities, digital technology still has the potential to contribute to mass harm. The technology has changed, but the human rights risks have not.