Skip to main content
Incident intelligence/SS-IR-012CASE FILE OPEN
Symbolic editorial illustration for SS-IR-012SERVANTSTACK // INCIDENT INTELLIGENCEFORENSIC IMAGE // VERIFIED FRAME
SS-IR-012 // INCIDENT REPORTReported

Facebook

Ranking Systems Amplified Anti-Rohingya Hate Speech During the Myanmar Genocide

EXECUTIVE BRIEF

For years Facebook's engagement-ranking systems pushed anti-Rohingya hate speech across Myanmar, where Facebook was effectively the entire internet.

SAFETY & SOCIETY
FAILURE CHAINTRACE COMPLETE
  1. 01TRIGGERFor years Facebook's engagement-ranking systems pushed anti-Rohingya hate speech across Myanmar, where Facebook was…
  2. 02MACHINE ACTIONDecision system
  3. 03MISSING GATERisk-based SME approval before execution
  4. 04IMPACTFinancial harm
01 // INCIDENT SUMMARY

The short version

For years Facebook's engagement-ranking systems pushed anti-Rohingya hate speech across Myanmar, where Facebook was effectively the entire internet.

02 // KEY FACTS

Case telemetry

INCIDENT
SS-IR-012
DATE
August 15, 2018
SYSTEM
Facebook
LOCATION / SCOPE
Myanmar (Rakhine State); moderation run from Kuala Lumpur
EVIDENCE
Reported
AI ROLE
Decision system
HARM
Financial harm
SOURCES
3 cited records
03ENTRY POINT // WHAT HAPPENED

The event

For years Facebook's engagement-ranking systems pushed anti-Rohingya hate speech across Myanmar, where Facebook was effectively the entire internet. On August 15, 2018 a Reuters investigation with UC Berkeley's Human Rights Center found more than 1,000 posts, images, and videos still live attacking the Rohingya as dogs, maggots, and rapists and urging they be exterminated; some had been online at least six years. On March 13, 2018 the UN Independent International Fact-Finding Mission concluded Facebook played a determining role in the violence, with chair Marzuki Darusman saying social media is Facebook and Facebook is social media in the country. The day the Reuters report ran, Facebook admitted it had been too slow.

04CAUSAL TRACE // AI'S ACTUAL ROLE

What the machine did

Facebook's recommendation systems were tuned to maximize engagement, and dehumanizing content is highly engaging, so the algorithm amplified genocidal hate speech at national scale with no local human review gate in front of it. Facebook had zero employees inside Myanmar and outsourced moderation to a secretive Kuala Lumpur operation called Project Honey Badger. In early 2015, as incitement reached tens of millions of users, only two reviewers could read Burmese. Algorithmic reach was effectively infinite; qualified human oversight was effectively zero.

Decision systemAutomation was a causal participant—not a decorative label for the system around it.
05BLAST RADIUS // CONSEQUENCES

Where the failure landed

The UN tied the amplification to an ethnic-cleansing campaign that drove more than 700,000 Rohingya into Bangladesh and left thousands dead amid killings, rape, and arson in Rakhine State. Facebook later admitted it did not do enough, removed Myanmar military accounts, and commissioned a human-rights assessment. In 2021 Rohingya refugees filed suits seeking more than 150 billion US dollars, alleging its algorithms promoted hate speech that fueled the genocide.

06 // EVIDENCE STATUS

Reported

Documented in the cited public record. Follow the sources for the precise evidentiary posture.

SOURCE RECORD UPDATED 2026-07-09

07 // SOURCE LEDGER

3 cited records

  1. 01
  2. 02
  3. 03
08CONTROL FAILURE // MISSING GOVERNANCE

Risk-based SME approval before execution

The failure pattern in this case: High-stakes output had no accountable checkpoint.

09INTERVENTION POINT // HUMAN IN THE MIDDLE

The moment the path could change

The appropriate subject-matter expert reviews the evidence, exceptions, and affected people before the output becomes action.

AI PROPOSES→HUMAN OWNS THE DECISION→SYSTEM EXECUTES
10CONTROL DEPLOYMENT // AUTHORITYGATE

Risk routing · named approval · audit trail

The AuthorityGate Operational Resilience framework requires a human SME validation gate keyed to each market a distribution system serves: a recommender cannot promote content in a language or region until certified, language-fluent human reviewer capacity for that region is provisioned and matched to projected volume. Under change-validation, scaling ranking into a country like Myanmar with tens of millions of users is a high-risk change that cannot be approved while only two Burmese-speaking reviewers exist; the gate blocks scale-up until human coverage is proportionate to reach, with a bounded-SLA SME escalation path so flagged incitement is not left live for six years.

RELEVANT KEYSTONE CONTROLHuman-in-the-Loop ValidationHow high-risk actions route to a named subject-matter expert who owns the go or no-go decision.
12 // THE ALTERNATIVE

Autonomy is a design choice.

See the operating model that keeps AI useful while preserving human authority at consequential moments.

Compare AgenticAI and AugmentedAI →