X Previews Labels Explaining Why AI Flagged Posts
X is previewing a system of labels that would tell users why its AI-powered moderation flagged their posts, according to a report from the outlet that covered the Brand & AI Safety Summit in New York. The labels would also let users appeal those decisions.
Designs shown at New York summit
Michael O’Herlihy, a Twitter veteran who recently rejoined X as head of safety, showed the designs on Tuesday, Nov. 10. The labels would appear beneath content on the platform. According to the report, they would explain what the AI detected in a post and describe how its reach was limited after it was flagged. They would also give users a way to appeal.
O’Herlihy suggested that X’s content moderation systems may display their decision-making process in the future. The designs were presented as a preview, not a launched feature, and no release timeline was given in the source material.
A mislabeled rocket launch
O’Herlihy’s pitch was grounded in a past error. X’s AI moderation system once mislabeled a post featuring an image of a SpaceX rocket launch as “adult content.” According to the report, nobody could see why it had been flagged, including the author of the post. O’Herlihy wants to fix that lack of transparency.
Auditable records versus human review
The report says O’Herlihy argues that AI-led moderation creates auditable records that are better than those produced by human review. Under that view, a system that logs and displays its reasoning is easier to check than individual human decisions.
Brand safety and advertisers
The announcement carries a business angle. The report notes that X’s advertising revenue has plummeted since Elon Musk took over the company, which makes brand safety a crucial pitch for winning advertisers back. Showing the summit’s audience of brand and AI safety professionals how moderation decisions are made and reviewed fits that effort.
For creators, the changes could matter if they reach users. Posts that are flagged or have limited reach often leave their authors without explanation. Labels that state the reason and the restriction, plus an appeal route, would address that gap. How the appeal process would work, and who would review appeals, was not detailed in the available text.
The full report is behind a subscription, and the excerpt available does not cover further details of the presentation. Further information on timing, scope or how the labels would work in practice has not been confirmed.
Source: adweek.com
Image: adweek.com
