The Facebook Files Reveal the Company's Own Research Found Its Algorithm Toxic for Teen Girls - and It Buried It

Identity verification · dual control

AuthorityGate's Operational Resilience framework requires a human SME change-validation gate before any engagement-optimization or recommender change affecting a high-risk population ships - and, just as importantly, a binding review gate that converts adverse internal safety findings into a mandatory hold. A qualified human Subject Matter Expert in adolescent mental health and product safety would have to review and sign off that an engagement-maximizing feed is acceptable for known minor accounts before it is deployed to them. When the company's own research found the algorithm worsened body image for one in three teen girls and was linked to suicidal ideation, that finding would have tripped a named human checkpoint with the authority to block continued deployment until the optimization target was remediated - not a slide deck that circulates internally while the unchanged algorithm keeps shipping. The failure here was structural: the harm was measured, but no accountable person was required to validate the recommender against that measurement before it kept running on children. The gate exists precisely so that "our own research said it was hurting them" becomes a stop condition, not a buried document.