Meta says a Claude tool got the No Kings Instagram account suspended

On Saturday, 17 October 2026, thousands of No Kings events took place across the United States. The coalition behind them used the weekend to push early voting ahead of the 3 November midterm elections, with marches to polling stations and help with voter registration. In the middle of all that, its official Instagram account disappeared.

A day later, Meta offered an explanation that turns a political dispute into an AI story. According to Meta's communications chief Andy Stone, the trigger was not a hacker and not a coordinated spam campaign. It was a Claude tool, reading Instagram profiles, posts and hashtags so fast that Meta's security systems decided they were looking at a bot.

Here is what happened, what Meta has and has not explained, and why this small incident says a lot about the next phase of the internet, where AI agents increasingly browse on behalf of people.

The three things you need to know

• The outage: On Saturday, Instagram took the No Kings Coalition's account offline during its nationwide day of action. The coalition says the page was unavailable for about 90 minutes, and its administrators were locked out even after the page was visible again. Full access came back at 5:33 p.m. New York time, according to the coalition.
• Meta's reason: In a post on X on Sunday, Andy Stone said Meta's network security systems had detected "the use of a Claude tool" reading profiles, posts and hashtags at a rate that sharply increased, about 4x the normal volume, in intense bursts. To Meta's automated protections, that looked like bot traffic.
• The open questions: Meta apologized and called the suspension an error, but did not say which Claude tool was involved, who was running it, or what it was doing with the account. The coalition says the explanation "still doesn't add up", and Senator Chris Murphy has called for an investigation.

What happened on Saturday

The No Kings Coalition is an alliance of hundreds of groups opposed to President Donald Trump's policies. The Next Web lists Indivisible, Common Cause and the ACLU among its members. Its latest day of action on 17 October was built around early voting. The coalition says nearly 3,000 events took place in all 50 states and in Washington, D.C., with millions of participants. Those are the organizers' own figures. Reuters describes rallies in cities from Seattle and San Diego to Chicago, Atlanta and New York.

For a movement like this, Instagram is not a side channel. According to Bloomberg, organizers called the platform a crucial way to talk to supporters. So when the account went dark during the rallies, the reaction was immediate. The coalition said Meta had pulled the account and locked out its admins. The profile came back after a while, but the people who run it still could not log in for some time.

Reuters puts the outage at more than an hour. The coalition says about 90 minutes for the public page, and longer for admin access, with everything working again at 5:33 p.m. Eastern time.

Meta's first explanation, and its second

Meta's story changed over the weekend, which is part of why critics are not satisfied.

On Saturday, Stone said the suspension was part of an automated response that sometimes happens when accounts post at very high frequencies. In other words, the first version pointed at unusually high rates of posting, sharing or commenting. He called the action "clearly" an error and apologized.

On Sunday, he gave a more detailed account. Meta's engineers had looked again and found that the spike was "primarily in the account's 'reading' behavior", he wrote. Not posting, but reading. Specifically, Meta's network security systems detected a Claude tool reading profiles, posts and hashtags, and the rate of that reading jumped to about four times the usual volume in intense bursts on 17 October.

Stone described those requests as abrupt, high volume and sequential, packed into a short period. That pattern, he said, resembled automated bot traffic rather than an ordinary person browsing, and it caused Meta's automated protections to take the account offline by mistake. An automated re-review about an hour later recognized that the removal should not have happened and restored the account. Sorting out the admin access, he added, took additional time.

What Meta did not say

Stone's statement is detailed about the pattern, but it leaves the most interesting parts open. Bloomberg points out that he did not specify which Claude tool was involved, who was operating it, or what activity it was carrying out in connection with the account.

"Claude tool" could mean many things. Anthropic's Claude can be used in a chat window, inside browser extensions, through agent features that click and scroll on websites, or through developer tools and the API that other companies build on. It could have been someone on the coalition's own team using an assistant to monitor reactions, a volunteer researching hashtags, or something else entirely. Meta's wording ties the activity to the account, but it does not explain how. None of the reports we read included a comment from Anthropic.

That gap matters, because without it nobody outside Meta can check whether the explanation fits the facts.

Why the coalition is not convinced

The No Kings Coalition did not accept the apology as the end of the story. On Saturday it said that restricting communications about voting in the middle of an election is unacceptable, and that Meta must answer for the suspension and make sure it never happens again, to any account.

On Monday, on Bluesky, the coalition went further. "Meta's explanation still doesn't add up," it wrote, and said it had received reports that No Kings and early voting content was inaccessible, restricted or seemed suppressed across Meta's platforms, beyond the Instagram account itself. Those are claims by the coalition, and Meta's response to them was not part of the reports we reviewed. The group's message was blunt: weeks before an election, an apology is not enough, and the public needs answers, accountability and concrete safeguards.

Democratic Senator Chris Murphy of Connecticut called for an investigation on Sunday evening, Business Insider reports. Even if the shutdown was an error, he wrote, it was a "total indictment" of Meta's outsized influence on culture and politics. Murphy has long pushed for stricter rules for big tech companies.

The coalition says its next step is a national organizing call on Wednesday focused on turnout before Election Day.

The real story: agents look like bots

Set the politics aside for a moment and look at the mechanism. According to Meta, a human-run account was taken down because software working for it read the platform faster than a human would. That is exactly the kind of collision we will see more often.

For more than a decade, platforms have built defenses against scrapers and bots on one simple assumption: humans are slow and irregular, machines are fast and systematic. A person scrolls, stops, reads a comment, goes back. A scraper fires off long runs of similar requests in seconds. Meta's description of the Claude traffic, abrupt, high volume, sequential, is the textbook pattern those defenses are trained to stop.

AI agents break that assumption. They act on behalf of a real person, often inside that person's logged in session, but they read and click at machine speed. From the platform's side, a helpful assistant summarizing a hashtag and a malicious scraper harvesting it can look almost the same. And when the defense fires, it does not hit the agent alone. It hits the account the agent is attached to, which in this case belonged to one of the most visible protest movements in the country, on its busiest day.

We have been tracking this tension for weeks. Big US banks have started blocking outside AI agents from logging in with customer credentials, as we reported in US banks block AI agents. OpenAI's new ChatGPT extension can read and control your open browser tabs, which we covered in ChatGPT's Chrome extension. And a recent test showed how persistent personal assistant agents can be when they chase a goal, in the impossible refund test. The No Kings case shows the other side: when agents act at scale, the systems around them react, sometimes clumsily.

Why it matters

There are three reasons this is bigger than a 90 minute outage.

First, timing and trust. An automated mistake that silences a political group during an election weekend is a serious event, even if nobody intended it. Platforms ask the public to trust that their enforcement is neutral. When the explanation arrives in two versions and leaves key facts out, that trust gets harder to keep, no matter what really happened.

Second, accountability for automation. Meta's story is that one automated system (bot detection) reacted to another automated system (a Claude tool), and a third automated system (the re-review) fixed it an hour later. In Meta's telling, people mostly came in afterwards, to restore admin access and to explain. As more of the web is run by software talking to software, the question of who is responsible when something goes wrong gets more complicated.

Third, the rules for agents are still missing. There is no common standard yet for how an AI agent should identify itself to a website, how fast it may read, or how a platform should treat an agent acting for a verified user. Until that exists, every platform will make its own guesses, and some of those guesses will hit the wrong people.

Dany's take

I believe Meta that its bot detection fired. Four times the normal reading volume in sharp bursts is exactly what those systems are built to catch. But I also understand why the coalition is angry. "A Claude tool" is not an explanation, it is a label. Which tool? Run by whom? Doing what? Meta has that data, and on a topic this sensitive, weeks before an election, it should share as much of it as it can. For me the bigger lesson is not about politics at all. We are all starting to give AI assistants access to our accounts so they can read, summarize and act for us. This story shows that the platforms are not ready for that yet, and neither are we. If your agent behaves like a bot, the platform will treat your account like one.

What this means for you

If you let an AI assistant or browser agent work with your social media accounts, think about speed and scope. Tools that read many profiles or hashtags in a short time can trip automated defenses, and the account that gets locked will be yours, not the tool's. Where a platform offers an official API or a partner tool, that is usually the safer route. Keep important accounts protected with more than one admin, and make sure you know how to reach the platform's support quickly if something goes wrong. If you manage an account for a group or a campaign, write down which tools have access and remove the ones you do not need. And if you build agents, give them sensible rate limits and let them identify themselves honestly. Should platforms treat AI agents like bots? Tell me in the comments under the video.

Sources

• Claude AI tool triggered No Kings Instagram suspension, says Meta (Bloomberg via Business Standard, 19 October 2026)
• Meta says it shut down No Kings Instagram account because it thought bots had been unleashed (Business Insider, 19 October 2026)
• Meta apologizes for suspending 'No Kings' Instagram account (Reuters via Arizona Daily Star, 19 October 2026)
• Meta says a Claude AI tool triggered the No Kings Instagram suspension (The Next Web, 19 October 2026)
• No Kings Coalition's official Instagram briefly went dark (The Verge, 17 and 18 October 2026)