Meta scales back outside moderation and hands content review to more advanced AI

Meta plans to significantly cut the role of outside contractors in moderating Facebook and Instagram. The company wants to hand more and more content review to its own AI systems, leaving humans with the riskiest and most contentious decisions.

How moderation is changing

Meta said that in the coming years it will gradually roll out more advanced AI systems across all its apps. This is not just spam filtering but full application of the rules to content: the company wants to find and remove severe violations faster and more accurately, including scams, illegal content, content related to drug sales, and other dangerous categories. If the new systems consistently outperform current methods, Meta will reduce its dependence on third-party contractors and strengthen its own internal teams and tools.

It is a notable shift for Meta's entire operating model. For years the company combined automated detection of violations with the work of thousands of outside moderators who manually reviewed disputed posts and reports. Now it is betting that modern models can better handle mass, repetitive and fast-changing work, where humans struggle to keep pace. For Meta, this is a question of speed, scale and the cost of moderation on platforms with billions of pieces of content.

What the AI can already do

The company says early tests of the new systems look strong enough to justify expanding their role. As evidence, Meta offers not general promises of better efficiency but concrete metrics on scams, fake accounts, sexual violations and suspicious ads. The company builds its case on these figures: part of the work previously done by outside vendors can now move to its own AI.

• Find and block about 5,000 additional scam attempts per day that existing review teams did not notice.
• Cut reports of accounts impersonating the most frequently imitated celebrities by more than 80%.
• Detect twice as much rule-violating content involving sexual solicitation aimed at adults, while making more than 60% fewer errors.
• Spot complex account takeover scenarios in which individual signals look harmless on their own but together point to an attack.
• After broader testing, cut views of ads containing scams and other serious violations by 7%.

Meta also stresses language coverage. According to the company, the new systems work in languages spoken by 98% of people online, whereas coverage was previously around 80 languages. The figure matters less than the fact that the models learn faster to adapt to local context: slang, emoji, code words and the behavior of niche communities. These are exactly the details that often make automated moderation a weak spot, especially when bad actors change tactics quickly.

Where humans will remain

Meta does not intend to remove humans from the chain entirely. The company says directly that AI will take on tasks better suited to the technology: repetitive reviews of graphic content, large-scale hunting for scam schemes, and categories where offenders constantly adapt to the rules and evade filters. But the most sensitive decisions will stay with people, such as appeals after an account is banned or cases that may lead to engagement with law enforcement.

Under this setup the human role changes rather than disappears. Experts will design, train, test and evaluate the systems, monitor bias and decision quality, and handle complex cases where context matters more than speed. Meta also promises that the community rules themselves are not changing because of this shift. At the same time, the company is launching an AI support assistant in Facebook and Instagram to help users understand why content was removed, how to file an appeal and what happens next.

What this means

Meta is moving AI from a supporting filter to the core of platform governance. If the transition works, the company will reduce its dependence on contractors and speed up moderation globally. But the cost of mistakes rises too: when a model rather than a human makes a disputed call, any systemic inaccuracy can instantly scale to millions of users. Meta's next test is to prove that this approach actually produces fewer errors, rather than simply cutting costs.

*Meta is designated an extremist organization and banned in Russia.