The British newspaper “The Guardian” revealed that Meta, the parent company of platforms like Facebook, Instagram, and WhatsApp, is struggling to manage content related to the Israeli-Palestinian war, particularly in Hebrew, despite recent changes to its internal policies. This information comes from new documents, including internal policy guidelines shared by a former Meta employee with the newspaper.
The documents disclosed that Meta does not have the same level of monitoring and oversight for Hebrew content as it does for Arabic content.
The former employee stated that “Meta’s policies governing hate speech related to Palestine are unfair,” a sentiment echoed by pro-Palestinian supporters.
Some employees on the front lines of the ongoing information battle surrounding the conflict reported feeling cautious about raising concerns for fear of retaliation, a worry reflected in a recent letter signed by over 200 Meta workers.
The former employee added, “These conditions give the impression that the company’s priorities do not genuinely involve ensuring that content is safe for the community.”
The documents, which became effective this spring, emerged as Meta and other social platforms faced criticism over their approach to content related to Palestine and the occupation, according to the newspaper.
In June, a coalition of 49 civil society organizations and several prominent Palestinians sent a letter to Meta accusing the company of “aiding and abetting genocide” through its content moderation policies.
Kat Knarr of the U.S. Campaign for Palestinian Rights, which organized the letter, said, “When Palestinian voices are silenced on Meta’s platforms, it has very real and direct consequences for the lives of Palestinians. People don’t hear about what’s happening in Palestine, but they do hear propaganda that dehumanizes Palestinians. The consequences are extremely severe and very real.”
“The Guardian” noted that content guidelines issued after the October 7th attack and the war in Gaza highlight a range of Meta’s content moderation policy decisions, including the company’s policies on hate speech and the Boycott, Divestment, and Sanctions (BDS) movement.
The policies require the removal of phrases like “boycott Jewish stores” and “boycott Muslim stores,” but they allow the phrase “boycott Arab stores,” according to internal documents.
According to the newspaper, the recent documents provide new insight into Meta’s ability to assess the quality of its content moderation in both Arabic and Hebrew.
Meta stated that it has “multiple systems in place to measure the accuracy of Hebrew content moderation, including assessments by Hebrew-speaking reviewers and auditors.”
The documents show that “there is no policy accuracy metric for Hebrew language implementation,” and the former employee explained that since Hebrew is not included in the system, this type of market-specific review in the Hebrew market is done on an improvised basis, unlike the Arabic market.
The former employee emphasized that “this discrepancy means the company reviews content in Israel’s official language less systematically than Palestinian language content.”
He added that the difference indicates a bias in how rules are applied to content, which may result in the over-application of rules to Arabic-language content.
Critics argue that due to the ongoing war, more attention is needed for Hebrew-language content, and Meta has faced questions before about its moderation of Hebrew-language posts, according to the newspaper.
Hebrew, spoken by around 10 million people, makes up a much smaller portion of posts on Meta’s social networks compared to Arabic, spoken by around 400 million people.
The newspaper noted that an independent analysis conducted in 2022, commissioned by the tech giant, concluded that its moderation system penalized Arabic speakers more than Hebrew speakers during the escalation of the Israel-Palestine conflict in 2021 – even when accounting for the disparity in the number of speakers.
This disparity is partly attributed to the fact that Meta, at that time, had implemented an “Arabic hostile speech classifier,” allowing for the automatic detection of violating content – such as hate speech and calls for violence – but did not do the same for Hebrew content, according to the newspaper.
Meta’s monitors say that the new documents reviewed by “The Guardian” show that even with the new Hebrew classifiers, less effort is being made to ensure the effectiveness of these latest measures, allowing such disparities in enforcement to persist.
In addition to hostile speech classifiers and other indicators, Meta uses sets of images, phrases, and videos that allow its machine learning tools to automatically flag and remove posts that violate its policies. Algorithmic moderators match content posted on social networks with materials in content banks, which have been previously judged to violate Meta’s rules.
The former employee mentioned that his colleagues noted that a number of images used in one of the banks created after the events of October 7th were mistakenly added, but there was no “process” to remove the images from this bank, leading once again to the potential over-application of content rules related to both sides of the Israeli-Palestinian conflict.
The former employee said Meta has responded to complaints from content moderators in the past. However, many employees expressed “fear of retaliation” or being labeled anti-Semitic if someone complained about the over-application of rules on pro-Palestinian content in Arabic, despite the company’s history, as they noted.
He added, “If I raise this issue directly, I feel my job would be at risk, and the company’s stance on this issue is clear.”
Meta’s reported revenue for the full year of 2023 was $134.9 billion. In its latest earnings report, the company posted total revenues of $39.07 billion for the second quarter of 2024.
