Particle.news
Download on the App Store
5 ARTICLES
·
4d ago

Meta to Automate 90% of Product Risk Reviews with AI

Meta’s Oversight Board warns that automation could undermine human rights protections

Up to 90% of risk assessments could become automated. Other companies like Klarna, Salesforce, and Duolingo are looking at job shifts in the face of AI.
Internal documents shed light on even more AI plans for Meta.
Internal documents shed light on even more AI plans for Meta.
Meta said it has noticed AI systems operating at a higher performance than humans for select policy areas

Overview

  • Meta’s AI will extend beyond “low-risk” releases to include AI safety checks, youth risk evaluations, misinformation detection and violent content moderation.
  • Product teams will submit questionnaires and receive near-instant AI-generated risk scores and mitigation recommendations to accelerate feature rollouts.
  • The company aims to phase out 90% of human-led risk assessments globally by 2025 while EU operations maintain human oversight under the Digital Services Act.
  • In April, Meta ended its human fact-checking program in favor of crowd-sourced Community Notes and algorithmic moderation tools.
  • CEO Mark Zuckerberg forecasts that within 12 to 18 months AI will write the majority of Meta’s own AI code, underscoring its reliance on automation.