Did a Claude Tool Get the No Kings Instagram Account Banned? What Meta Said, What It Didn't, and What It Means for Anyone Running AI Agents on a Social Account

What happened on October 17, hour by hour

The day was the coalition's national early-voting push ahead of the November 3 midterms, with what the organizers count as nearly 3,000 events across the country. Times below are Eastern, and the source for each is noted, because the accounts disagree by about half an hour on the start.

What Meta actually said, word for word

Two statements, a day apart, both from Andy Stone, whose X bio reads 'Communications, @Meta' (Bloomberg calls him vice president of communications). Both were cross-posted to Threads. There is no Meta Newsroom post and no outlet quotes a separate on-record statement with new facts, so this is the entire official record.

The coalition pushed back on Threads within hours: 'In the 24 hours before our account was shut down, we posted only a handful of times and livestreamed to thousands. Meta's "high-frequency" excuse doesn't add up.' Sunday afternoon, Stone came back with a different mechanism.

Read it closely and three things stand out. First, the two explanations are not the same: Saturday's was about the account posting, sharing, or commenting too often; Sunday's is about the account reading too much, and Stone's own word is 'primarily,' not 'entirely.' Second, the phrase is 'a Claude tool,' not 'Anthropic's Claude,' not 'an AI agent,' and not a product name; the 'Anthropic PBC's Claude' wording in circulation is Bloomberg's paraphrase. Third, 'the account's reading behavior' implies the tool was acting as, or logged into, @nokingscoalition, which would make it something the coalition or a vendor was running, but Stone never says that, and the coalition's Sunday-night reply does not address the Claude claim at all. Anthropic has not commented publicly as of October 19; none of the outlets covering the story report reaching the company. We will update this post if either side says more.

What 'a Claude tool' could mean

Nobody has identified the product, so what follows is a map of the possibilities, not a finding. There are four ways something called Claude reads Instagram profiles, posts, and hashtags in bursts, and they have very different implications.

• Claude in Chrome. Anthropic's browser extension, available on Pro, Team, and Enterprise plans since December 2025, drives the user's own logged-in Chrome tab. Ask it to 'go through the hashtag and note which creators posted today' and it will open profile after profile, fast, under the account's own cookies. That is exactly the signature Stone describes, and it is the most likely reading if the account's own team was using Claude. Anthropic's help pages for the extension say nothing about request rates or respecting a site's terms.
• Claude with computer use or a desktop agent. Same idea, screenshots and clicks instead of a browser extension. Also runs as the logged-in user, also bursty.
• A custom agent built on the Claude API. A campaign vendor's monitoring or social-listening tool that uses Claude for the thinking and a headless browser for the reading. Meta's security systems can sometimes fingerprint the browser automation, but they would have no way to know Claude was the model behind it, which makes this the hardest to square with Stone naming Claude at all, unless the tool identified itself.
• Claude Code or a scripted job. A developer-side script that hits Instagram's web endpoints. Possible, but the least likely thing a communications team is running on a protest day.

The question that decides which of these is true, and whether the coalition bears any of the responsibility Meta is implicitly assigning it, is simple: was the Claude tool logged into @nokingscoalition? Meta's wording says yes. The coalition hasn't said. Until one of them does, 'Claude got No Kings banned' is a headline about Meta's claim, not an established fact.

Does Meta's explanation add up?

Parts of it are plausible on their face. Instagram does run automated anti-scraping enforcement, and a logged-in session suddenly reading four times its usual volume in tight bursts is the textbook pattern those systems look for. An automated re-review an hour later is also how Meta's enforcement is designed to work. But the record also contains things Meta has not explained.

• The story changed. Saturday it was posting frequency; Sunday it was reading volume. A company that got the mechanism wrong on day one, in writing, is asking to be taken at its word on day two.
• Admin access took four hours, not one. Meta's statement handles this with 'it unfortunately took additional time.' For an organizing account on its biggest day of the fall, the gap between 'restored' and 'usable' is the whole story, and it has no technical explanation yet.
• Other accounts. Levin reported pro-No Kings creators suspended the same day; the coalition says it has reports of #NoKings and early-voting content restricted across Meta platforms. These are organizer claims, unverified, and Meta hasn't addressed them, but the Claude explanation covers one account and only one.
• No product, no operator, no Anthropic. Meta named a competitor's AI model in a politically charged takedown and gave no detail that would let anyone check it. Anthropic, which usually responds quickly when Claude is in a headline, has been silent for a day.

Sen. Chris Van Hollen's Saturday-night reaction, 'Censoring free speech to appease Donald Trump? We the people deserve answers,' captures the suspicion Meta is now arguing against. The Claude explanation, if true, is a better story for Meta than a political one. That is also a reason to want it confirmed by someone other than Meta.

Can an AI agent get your Instagram account suspended? Yes, and here's the rule

Whatever happened to No Kings, the policy question has a clear answer, and it matters for every brand, campaign, and creator now pointing agents at their own accounts. Instagram's Terms of Use, section 4.2, say: 'You can't attempt to create accounts or access or collect information in unauthorized ways. This includes creating accounts or accessing or collecting information in an automated way without our express permission, regardless of whether such automated access or collection is undertaken while logged-in to an Instagram account.' The last clause is the one that bites: being logged in as yourself is not permission. Instagram's help page on data-scraping restrictions lists 'You used an app or service to interact with Instagram in unauthorized ways' as a trigger, and Meta's separate Automated Data Collection Terms define the problem as 'the collection of data from Meta Company Products via the use of automated or programmatic tools,' naming 'web scrapers, bots, robots, spiders, crawlers, user-agents.' A browser agent reading forty profiles in a minute is, by that definition, a user-agent collecting data. Meta has published nothing specific to AI browsing agents, so the general rule is the rule.

On the Anthropic side, the Usage Policy prohibits using Claude to 'circumvent the guardrails or terms of other platforms or services' and to 'utilize automation in account creation or to engage in spammy behavior,' which puts the burden on the person directing the agent. Nothing in Anthropic's Claude in Chrome documentation throttles the extension or warns that a logged-in burst can look like a bot. That is a gap worth closing, because the user experience is 'ask Claude to check the hashtag' and the platform's experience is 'this account is scraping.'

If you run agents on an account you can't afford to lose

• Separate the reading from the account. Do research and monitoring from a session that isn't logged into your main account, or through Meta's official APIs where your use case qualifies. A burst that gets a throwaway session rate-limited is annoying; the same burst on your verified account is a takedown.
• Slow the agent down on purpose. Ask for one profile at a time with a pause, or cap it at a handful per task. Agents are fast by default and have no idea what 'four times normal' looks like for your account.
• Never run it on the big day. Whatever No Kings was or wasn't doing, the enforcement landed at peak mobilization. Do the research the day before.
• Keep a second admin and a recovery path. The coalition's account was visible again in about ninety minutes and unusable for four hours. Make sure someone else can post, and know where Meta's appeal flow is before you need it.
• Assume Meta can see the tool. Stone's post says the security systems 'detected the use of a Claude tool.' Whether that was a fingerprint or an identifier, plan as if the platform knows what's driving the browser.

Our read

This is the first time a major platform has publicly blamed a named AI assistant's browsing for taking a political account offline, and it arrived with less evidence than the claim deserves. Meta's second explanation is specific enough to be checkable and vague enough that no one can check it. The coalition has answered the political question loudly and the technical one not at all; if its team or a vendor was running Claude in Chrome against Instagram on October 17, saying so would cost it little and settle a lot. And Anthropic's silence is the odd one: its model was named, in public, by Meta, as the proximate cause of a takedown during an election, and a short statement about what Claude in Chrome does and doesn't do on sites like Instagram would help every user, not just this one. For readers, the durable lesson has nothing to do with No Kings: the agents in our personal AI agents ranking and the browsing features in Claude are built to act as you, and the platforms they act on have rules written for a world where 'you' was a person clicking at human speed. Until the platforms and the agent makers agree on what an agent may do while logged in, the account holder carries the risk. Related: Can Claude ban you for being mean to it? and Can Claude and ChatGPT report you to the police? We will update this post if Meta names the tool, if No Kings addresses it, or if Anthropic comments.