Meta Shifts Toward AI-Automated Oversight Across Facebook and Instagram

Meta is expanding its use of artificial intelligence to guide decisions across Facebook and Instagram, with internal systems increasingly responsible for coding, ad targeting, and now, user safety and product risk evaluations.

According to a recent NPR report, Meta plans to automate up to 90% of its product risk assessments – evaluations traditionally handled by human teams. These assessments consider factors such as user privacy, child safety, and the potential spread of harmful or misleading content. While Meta has assured that final oversight will remain human-led and only low-risk decisions are being delegated to AI, the scale of automation signals a shift in how the company intends to manage integrity reviews going forward.

Meta’s Q1 2025 Transparency Report further detailed changes to its enforcement strategy. The company has revised how it handles “less severe” policy violations by adjusting its confidence thresholds for automated takedowns. According to the report, Meta now requires stronger evidence before removing content and is eliminating many forms of content demotion. If automated systems are found to be underperforming, Meta says it deactivates them entirely to recalibrate.

This new approach has led to fewer moderation errors – Meta reports a 50% drop in enforcement mistakes. However, the flip side is that more harmful content may be slipping through. In Q1, automated detection of bullying and harassment on Facebook fell by 12%, indicating that fewer violative posts were flagged by the system.

CEO Mark Zuckerberg has suggested that AI will soon be responsible for much of Meta’s code base, further signaling a company-wide reliance on automation. As Meta leans into AI for decision-making at scale, questions remain about the risks of reducing human oversight in areas with significant implications for user experience and safety.