Meta Says a Claude Tool's 4x Reading Spike Tripped Its Bot Detector and Locked No Kings Out of Instagram

Meta says its bot detector locked the No Kings Instagram account on Saturday because it saw a Claude tool reading profiles, posts and hashtags about four times faster than usual. The system read that as a bot. The group was running a nearly 3,000-rally early-voting push that day.

What Meta says happened

Stone first called the shutdown an incorrect enforcement triggered automatically. A day later he posted a fuller account on X. Per Business Insider and Mother Jones, which both quoted it, the spike was "primarily in the account's 'reading' behavior." Meta's systems saw a Claude tool pulling profiles, posts and hashtags in intense bursts, and the pattern resembled automated traffic.

The timeline is where the two accounts split. Stone said an automated re-review about an hour later recognized the removal was wrong and restored the account. He also acknowledged it "took additional time" to sort out admin access. No Kings, as Business Insider reported, says administrators were locked out for several hours during the rallies. Both can be true, and Meta has not published a clock-time log to settle it.

What Meta has not said

Meta's public explanation, as quoted by the outlets I could read, does not say who was operating the Claude tool. It does not say whether the group's own staff or a volunteer was using an assistant to read the account's feed, or whether something else was pointed at it. Anthropic's comment is not in any of the coverage I found. Bloomberg's report was paywalled beyond its summary, so I could not check its full text.

That gap matters because the fix depends on it. If a staffer ran an AI assistant to track hashtags during a rally weekend, that is ordinary use that Meta's detector punished. If the activity came from somewhere else, the account was a target or a bystander. The stated cause, a rate spike, fits either story.

Why a rate limit looks political on a rally weekend

A false positive on a bot filter is routine. A false positive on October 17, with early voting under way and the Nov. 3 midterms ahead, lands differently. Connecticut Sen. Chris Murphy, who has pushed for stricter antitrust limits on Big Tech, called for an investigation. Even if it was an error, he wrote, it was a "total indictment of Meta's outsized influence." His second theory, that Meta is courting Trump and so skipped the prep work, is a motive claim. Nobody in the coverage offers evidence for it.

No Kings put it in operational terms, saying that restricting voter communications mid-election is "unacceptable" and that Meta "must answer" for it. Mother Jones lined the episode up with a TikTok suspension of the Musk documentary's account, restored Sunday as an "error," and Brazilian complaints about Lula's Instagram posts. Meta denies favoring any side in Brazil. Those are separate incidents with separate stated causes, and one shared label does not make them one pattern.

I could not find a base rate for how often Meta's automated systems wrongly lock accounts of this size, so there is no way to say whether a few-hour outage is unusual or typical. Meta has not published one.

The product question underneath

Here is the product question. An assistant that reads a social account on a person's behalf produces traffic that can look like a scraper, because it fetches pages faster than a human scrolls. Whether Meta's rules allow that depends on its terms, which the coverage does not quote and which I have not verified.

Meta has said the removal was wrong and apologized. It has not said whether it will change how the detector treats AI-assisted reading, and nothing in the coverage says it already has. That is the open item, and it is a spec question Meta can answer in one paragraph.