Facebook knew it was being used to incite Ethiopia violence, records show

Facebook used by militias ‘to seed calls for violence’

CNN’s publication of these warnings from within Facebook comes seven months after a Facebook team initially shared an internal report entitled “Coordinated Social Harm.”

Ethiopia is an ethnically and religiously diverse nation of about 110 million people who speak scores of languages. Its two largest ethnic groups, the Oromo and Amhara, make up more than 60% of the population. The Tigrayans, the third largest, are around 7%.

One of the groups flagged in the March report was the “Fano,” an ethnic Amhara militia group with a reputation for brutality that has been drawn into the war in Tigray, sometimes fighting alongside Ethiopian government forces. Facebook said it had observed a cluster of accounts affiliated with the militia group, including some based in Sudan, using its platform to “seed calls for violence,” promote armed conflict, recruit and fundraise.

Though the Facebook team said it had recommended the Fano-affiliated network be taken down, it suggested that other bad actors promoting violence on its platform were simultaneously slipping through the cracks. In a headline in bold, the team warned: “Current mitigation strategies are not enough.”

The Facebook documents also detail the platform’s removal of a cluster of accounts linked to the Oromo diaspora, mostly based in Egypt, which was targeting Ethiopian audiences with highly inflammatory content, including “explicit calls to violence against government officials and other ethnic groups.” One inciteful post highlighted in a report shared a photo of what appears to be a Molotov cocktail being lit and the statement: “Burn the whole country down.”

The whistleblower, Haugen, said one of her core motivations for gathering the internal documents was bringing to light “how badly Facebook is handling places like Ethiopia,” where she suggested engagement-based ranking was fanning ethnic violence.

“I genuinely fear that a huge number of people are going to die in the next five to ten years, or twenty years, because of choices and underfunding” by Facebook, Haugen said.

In comments made to the consortium, Haugen emphasized the vast difference between the integrity and security systems rolled out by Facebook in the United States versus the rest of the world, adding that the company was not adequately policing its platform in most non-English languages.

“The raw version [of Facebook] roaming wild in most of the world doesn’t have any of the things that make it kind of palatable in the United States, and I genuinely think there’s a lot of lives on the line — that Myanmar and Ethiopia are like the opening chapter,” she said.

A Facebook spokesperson told CNN that the company has been “actively focused on Ethiopia.”

“Over the past two years we have actively invested to add more staff with local expertise, operational resources and additional review capacity to expand the number of local languages we support to include Amharic, Oromo, Somali and Tigrinya. We have worked to improve our proactive detection so that we can remove more harmful content at scale. We have also partnered extensively with international and local experts to better understand and mitigate the biggest risks on the platform,” the spokesperson said.

Current mitigation strategies are not enough

None of the revelations from the Facebook documents are news to activists and human rights groups, who have warned for years that the social media giant has made insufficient efforts to protect human rights in Ethiopia, Africa’s second most populous country.

But as Ethiopians began to use Facebook to engage in public debate, observers saw that the platform was being abused by a variety of actors, including politicians, to incite discrimination and violence.

“Given the experience in Myanmar, it was really incumbent on Facebook to do a human rights impact assessment and evaluate what they needed to do so that Facebook in Ethiopia didn’t become a place for incitement to violence,” Kaye said, adding that he didn’t know what that assessment looked like or if it was done.

In June 2020, a Facebook employee posted a report to an internal group with about 1,500 members called “At Risk Countries FYI” recapping an ongoing audit into how well its Artificial Intelligence and other signals, like third party fact checkers, worked in the most at-risk countries where the platform operates.

“We found significant gaps in our coverage (especially in Myanmar & Ethiopia), showcasing that our current signals may be inadequate,” the employee wrote, sharing a spreadsheet with a list of at-risk countries and the languages supported by the platform in each.

The spreadsheet showed that Facebook had failed to build automated systems, called classifiers, to detect misinformation or hate speech in Oromo or Amharic — two of the most widely spoken languages spoken in Ethiopia.

Even as the conflict in Tigray escalated, Haugen said she had only found evidence that Facebook had allocated “even slight language support” in two of the country’s many native languages.

PesaCheck has five full-time Ethiopian fact-checkers working in four languages — Oromo, Amharic, Tigrinya and English — but says it recently had to relocate one staff member from Ethiopia due to intimidation. AFP Fact Check employs one fact-checker in Ethiopia, Amanuel Neguede, who reviews content in Amharic and English.

Each day, Neguede told CNN that he reviews thousands of posts on an internal Facebook tool, which surfaces content flagged as false or misleading through a combination of AI and human moderators. Originally, Neguede said that AFP only had access to English-language content in Ethiopia through the tool, which would surface only limited content each day.

The tool began serving AFP Amharic-language content in May, and now the number of claims he says he sees on a daily basis has drastically risen. The tool does not always accurately identify mis- and dis-information, but Neguede says it helps with his work.

“I’ve seen a lot of a lot of hate speech, that definitely does fuel ethnic violence in Ethiopia,” Neguede said.

“Whenever there’s a major offensive, for example that’s happening in the north, we can see a lot of images of conflict that’s happened in different countries used in a misleading context. I would say that most of the time we’ll see posts surface — especially posts that are widely shared — after real news events.”

For that reason, Taye, now an independent Nairobi-based analyst, works with grassroots volunteers to collate misinformation and hate speech they spot on the platform into Excel spreadsheets, which they then send on to Facebook for removal. But she says that much of what they flag — including posts calling for the extermination of certain ethnic groups — does not get taken down and, occasionally, the company has responded to activists asking for them to translate posts.

Facebook says it has improved its reporting tools for people in Ethiopia to make it faster and easier for them to report content they believe violates its community standards, which are now available in Amharic and recently launched in Oromo. It says it has also established reporting channels for international and local human rights groups and civil society organizations to flag potentially harmful content in Ethiopia.

Content moderation is extremely dangerous work in Ethiopia, Taye says, adding that she has personally been accused of siding with a person or group because a post was removed. The task is also mentally exhausting; volunteers spend hours reviewing graphic content, often doing it alongside full-time jobs and raising their kids. Taye said it was unacceptable that poor people have effectively been left to the “dirty work” of one of the world’s richest companies.

Facebook will not reveal exactly how many local language speakers are evaluating content in Ethiopia that has been flagged as possibly violating its standards, or how much it has invested in resources to better police its platform in the country. A Facebook spokesperson said the company had invested “$13 billion and have 40,000 people working on the safety and security on our platform, including 15,000 people who review content in more than 70 languages working in more than 20 locations all across the world to support our community. Our third party fact-checking program includes over 80 partners who review content in over 60 languages, and 70 of those fact checkers are outside of the US.”

“Our track record shows that we crack down on abuse outside the US with the same intensity that we apply in the US,” the spokesperson added.

A war on social media

Since war broke out in Tigray, supporters of both the government and TPLF have waged a parallel fight on social media, creating a virtual battleground of toxic ethnic and religious hatred. But it’s difficult to determine when atrocities unfolding on the ground are the direct result of hateful content shared online.

Activists say that divisive language has been echoed by Abiy’s supporters online.

CNN has reached out to the Ethiopian prime minister’s office for comment.

A Facebook page with an INSA-affiliated administrator was also flagged in internal Facebook documents in March, but was not recommended for removal.

As the conflict escalates online and off, the Ethiopian government has accused Facebook of blocking user and removing posts “disseminating the true reality about Ethiopia.”