Meta pushes for better content enforcement with more advanced AI

Tech giant Meta plans to improve content enforcement with more advanced AI. The aim is to better detect and remove harmful posts, scams, and illegal material before they spread widely.

The company explained that its new systems would catch severe violations more accurately, respond faster to real-world threats, and reduce over-enforcement which results in legitimate content being taken down wrongly.

Meta said this is part of a broader plan to reduce reliance on third-party vendors and strengthen its own internal systems and workforce.

On the enforcement side, Meta noted that the real advantage lies in scale. It has been testing AI systems that can spot patterns people might miss, including suspicious login activity, celebrity impersonation, adult sexual solicitation, and scam websites.

According to the company, early tests identified and mitigated around 5,000 scam attempts a day that no existing review teams had caught. They also reduced reports involving the most impersonated celebrities by more than 80%, and cut errors in detecting adult sexual solicitation by over 60%.

Meta also said the systems can operate in languages spoken by 98% of people online, compared to roughly 80 languages previously. This is significant because harmful content often evolves quickly and appears in local slang, emojis, or coded language that is harder to track manually.

Meta’s third quarter 2025 integrity reports, published in December 2025, outlined the scale of enforcement challenges across its platforms. The reports highlighted continued reliance on automated systems to detect violations, alongside human review for more complex decisions.

While human reviewers are not being removed, Meta wants AI to handle repetitive and fast-changing tasks, while people focus on higher risk decisions such as appeals and law enforcement referrals. This approach can improve speed and coverage, but it also raises concerns about errors at scale. Meta said it is testing safeguards to improve accuracy and reduce bias.

Alongside these enforcement changes, Meta is also expanding how users access support. The company previewed its Meta AI support assistant and began testing a new support hub on Facebook and Instagram. In March, it started rolling out the assistant globally across apps and on desktop Help Centres.

Meta said the support assistant can help users with account issues such as password resets, privacy settings, and profile updates, and with reporting scams or impersonation accounts. It is designed to provide 24/7 assistance, often within seconds, and is available in all languages supported by Facebook and Instagram for support topics, the company said.

The company is also testing its use for login assistance in the United States and Canada, with plans to expand further.

Earlier this month, Meta acquired Moltbook, a social network where only software programmes can post and interact.

The company spent $72.22 billion on capital expenditure in 2025. This will rise to between $115 billion and $135 billion in 2026, said Chief Financial Officer Susan Li, during Meta’s Q4 2025 earnings call, driven by increased investment in Superintelligence Labs and its core business.