Meta employees are reportedly warning that the company’s AI moderation rollout is advancing too quickly. The Decoder reports that large language models have already replaced about half of human moderation requests, with plans to exceed 90% for certain content types.
The concern is that moderation errors can have major consequences for speech, safety, and platform trust. Replacing human review at scale requires strong evaluation, escalation paths, and transparency around failure rates.
The story shows the operational risk of using LLMs in high-volume decision systems where mistakes affect real users.