The Algorithm That Boosted a Junta
Weeks after Myanmar's military seized power in 2021, and after Facebook barred the armed forces from its platforms, a rights group found the company's recommendation system was still steering users toward pro-military pages. Several of those pages carried content that broke Facebook's own rules on violence and misinformation.
On Feb. 1, 2021, Myanmar's military seized power in a coup, detaining the country's elected leaders and triggering mass protests that security forces met with lethal force. Three weeks later, on Feb. 24, Facebook said it was banning the military, known as the Tatmadaw, along with military-controlled state and media entities, from Facebook and Instagram, citing the risk that “online threats could lead to offline harm,” according to the company's announcement.
Yet the ban did not stop the platform's own machinery from promoting pro-military content. In a report published on June 22, 2021, the campaign group Global Witness said Facebook's page-recommendation algorithm was still directing users to material that violated the platform's policies on violence and misinformation, according to The Associated Press.
A clean account, five recommendations
To test the system, Global Witness set up a new Facebook account with no history on March 23, 2021 — shortly before the deadliest stretch of the crackdown — and searched for “Tatmadaw,” the Burmese name for the armed forces. After the account clicked “like” on a military fan page, Facebook offered a set of “related pages.” The group liked the first five “related pages” that Facebook suggested; three of those five pages contained content that violated Facebook's policies, according to TIME.
Among the material surfaced was a March 1 post on one recommended page that threatened protesters, telling those who damaged surveillance cameras and cut cables that soldiers had “been given an order to shoot to kill them on the spot,” according to Global Witness's findings. With scores of demonstrators already shot dead, the group said, such threats were credible.
Enforcement behind the algorithm
The finding pointed to a gap between what Facebook said it was doing and what its systems were doing. “The platform operates too much like a walled garden, its algorithms are designed, trained, and tweaked without adequate oversight or regulation,” Naomi Hirst, head of the digital threats campaign at Global Witness, said in remarks reported by The Associated Press.
Sophie Zhang, a former Facebook data scientist who became a whistleblower, told the AP: “Once again, Facebook shows that it's good at making broad sweeping announcements and bad at actually enforcing them,” the news agency reported.
Facebook's response
Facebook said its teams “continue to closely monitor the situation in Myanmar in real-time and take action on any posts, Pages or Groups that break our rules,” according to The Associated Press. The company has said it proactively detects most of the hate speech it removes in Myanmar and that banning the Tatmadaw made it harder to misuse its services.
The episode underscored a recurring pattern researchers describe with engagement-ranked feeds: recommendation systems that reward provocative material can outpace a company's ability to enforce its own rules, particularly in high-risk moments and in languages where platforms invest less in moderation. TIME reported that Facebook had tried to ban Myanmar's military even as its own algorithm kept promoting pages supporting it.