Meta fails to meet ‘gold standard’ of hate speech removal
An expert confirmed that hateful phrases like “White people are all Nazis,” “Trans people don’t exist,” and “Black people are more violent than white people” are just a few of the sentiments allowed under Meta’s dehumanizing speech policy.
In a written statement to the Royal Commission on Anti-Semitism and Social Cohesion, the company argued that this was not their “role in police aggression”.
“We also allow statements that appeal to common morality (such as ‘immigrants are criminals’) or comparisons to known hate groups or criminal organizations (such as ‘All white people are Nazis’),” the statement said.
When it comes to content that violates its policies, the commission heard Meta did not meet the “gold standard”, citing fears of over-enforcement.
Benjamin Good, Global Director of Meta’s Core Policy Team, said the company “has no intention of continuing infringing behavior” on Facebook, Instagram and Threads.
Speaking at the inquiry, Mr Good said a proactive approach was the “gold standard in removing infringing content before it is displayed” but the method “carries risks”.
“When we proactively remove content, … if we are wrong, if the content is not infringing, then there is a significant risk of overenforcement, which is what we discussed when we made the changes in January last year.”
In early 2025, Meta shifted to a largely reactive approach to standards, relying on users to report content, which is then analyzed by AI or human moderation for validity.
“For less serious policy violations, we will rely on someone reporting the issue before taking any action,” a Meta statement said.
“This means we’ll proactively catch less bad stuff, but we’ll also reduce the number of innocent people’s posts and accounts we accidentally remove in the process.”
When Mark Zuckerberg announced sweeping content moderation changes, he described the policy change as a “change.”
Before the changes, one in ten content removed was reportedly innocent material, with Meta arguing that this meant “too many mistakes and too much content being censored”.
Mr Good said the reactive approach was “a really important part of the toolset”, allowing for more accurate removal of content.
“For example, in times of crisis we have heard from Jewish communities and others that their content has been removed when they were trying to speak out against atrocities or condemn a terrorist group like Hamas, for example,” he told the commission.
The 2025 Meta Enforcement first quarter report found a 50 percent reduction in app errors from 2024 to 2025, claiming that “low prevalence of infringing content remains largely unchanged in most problem areas” during that time.
But the change in policy resulted in a 79 per cent reduction in action taken against harmful content on Facebook, from 5.8 million items responded to in October-December 2024 to 1.2 million between July-September 2025, the royal commission found.
More to come.
