Meta moves content moderation to in-house AI, away from third-party vendors

Meta has announced new AI systems for moderation and enforcement on its platforms: Facebook, Instagram and Threads. At the same time, the company is cutting its dependence on third-party contractors, who previously handled a significant share of the work of identifying violations. According to Meta, its in-house AI tools outperform earlier approaches on several key measures.

The new systems detect violations of platform rules more accurately, block fraudulent schemes more effectively, respond faster to real-time events and trigger fewer false positives, meaning they are less likely to remove legitimate content by mistake. The move fits into a broad transformation Meta has been carrying out since early 2025. In January, the company announced it was ending its independent fact-checking program in the US and switching to a Community Notes model, with user-written explanatory notes similar to X's.

Meta also relaxed a number of policies on political content at that time and adjusted how its algorithmic recommendations work. Moving from third-party vendors to in-house systems is a logical continuation of that course. External moderation contractors have always been a weak link: they worked more slowly, cost more and had no direct access to the platform's internal signals.

Those signals are what allow Meta's systems to spot patterns of violations much earlier than a team of external reviewers can. Speed of response to real-world events is one of the critical parameters for platforms with audiences in the billions. During natural disasters, elections, military conflicts and other crises, the volume of potentially harmful content rises sharply.

AI systems with constant access to real-time data streams objectively have the edge over external review teams in exactly these peak situations. The stated goal of reducing overenforcement, meaning excessive blocking, deserves separate attention. It is a long-standing pain point for Meta: removals of news content, journalistic publications, medical materials and statements by minorities have repeatedly drawn criticism of the company.

According to Meta's representatives, the new systems should draw a sharper line between real violations and borderline cases. At the same time, a full shift to automated moderation carries its own risks. AI systems are trained on historical data and can reproduce biases embedded in their training sets.

Without a transparent appeals mechanism and independent audit, users and regulators will not be able to judge how fairly these systems work in practice. European regulators already require large platforms to disclose information about their moderation systems and to allow appeals of decisions under the Digital Services Act (DSA). There is no US equivalent of these requirements yet, but regulatory pressure around the world continues to grow.

For Meta, moving to its own AI systems is both a technological and a reputational choice. Technologically, the company gets a more flexible and scalable tool that can be adapted and fine-tuned for specific tasks. Reputationally, it shows that it is taking direct control of key platform infrastructure rather than outsourcing it.

Meta has not yet publicly disclosed how exactly the new systems are built, which models underpin them or how human oversight is organized. The answers to those questions will determine whether the updated approach is a real step forward for content safety or just another round of opaque governance of the world's largest social platforms.

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