CLEARRULES · OPINION
Harassment does not fit into one report
The proposal to connect abuse reports with AI deserves attention—not because a machine should decide guilt, but because people should not have to explain the whole history again every time.
On 2 November 2026, Oversight Board member Helle Thorning-Schmidt proposed using AI assistants to connect harassment reports and escalate serious cases. This is a proposal in an opinion article, not an announcement of a new Facebook or Instagram feature.[1]
The important word is less “AI” than “connect.” A person experiences harassment as one continuing story. A service may receive it as separate reports. Between those two accounts, something crucial can disappear: why this particular new post is frightening now.
What changed after the second review
Thorning-Schmidt returns to the case of a journalist. In its 4 June decision, the Board described how her complaints to Meta employees she knew led to two internal escalations. After the second review, the company permanently disabled the Instagram account posting threats. The Board upheld that decision.[2]
That result matters: telling the story as if the company never intervened would be wrong. But the Board also condemned the delay in reviewing the threats. Successful intervention through personal connections does not explain how someone without contacts inside the company can receive the same attention.[2]
An assistant should preserve facts, not strengthen an accusation
Imagine someone has reported several related posts and another appears. A useful assistant could retain links to the earlier reports and show the reviewer the sequence of events. This is a hypothetical system design, not instructions for an existing Meta tool.
The help would not consist of writing a more persuasive complaint. It would consist of not losing information already supplied. Otherwise the person spends energy repeating the account, while each new reviewer receives another fragment of the story.
Connections between reports need reasons too. Several people may independently criticize the same speech; a shared topic does not establish coordinated harassment. If an assistant groups posts, a reviewer should be able to understand why.
Connecting reports need not mean automatic bans
The Board’s June decision protects more than the person being threatened. It also insists on understandable reasons for disabling accounts and fair appeals. That is not a concession to harassment: examining context should help distinguish abuse from legitimate speech, not remove the need to distinguish them.[2]
For ClearRules, the useful test for such an assistant is one question: can the next reviewer see the relevant history without making the person targeted assemble it all again? The answer is not measured by the elegance of generated prose.
The proposal deserves support precisely as a way to take unnecessary work off someone already under strain. Information can be brought together faster. A sanction must still rest on checked circumstances and come with an explanation.