Meta now says an Anthropic Claude tool was behind the mistaken suspension of the No Kings Coalition's Instagram account on Saturday, October 17. The account went offline in the middle of the group's nationwide early-voting events ahead of the November 3 midterms. Meta's revised explanation is short, and it puts a named AI agent product at the center of an anti-bot enforcement error.
What Meta said
Andy Stone, Meta's vice president of communications, gave two explanations on X, a day apart.
The first, posted Saturday at 3:07 p.m. PT, said the account was caught by automated systems that fire "when accounts post, share or comment at very high frequencies." He called the action "clearly taken in error" and apologized (Stone, October 17).
The second, posted Sunday at 1:41 p.m. PT, changed the cause. "Our engineering team discovered that the spike that triggered the incorrect enforcement was primarily in the account's 'reading' behavior," Stone wrote. "Specifically, our network security systems detected the use of a Claude tool reading profiles, posts, and hashtags, the rate of which sharply increased (~4x volume) in intense bursts on October 17." He added that the "abrupt, high-volume, sequential requests in a short period resembled automated bot traffic rather than an ordinary human browsing our systems" (Stone, October 18).
According to Stone, an automated re-review about an hour after the removal restored the account, but it "took additional time" to sort out admin access. The coalition said its page was publicly unavailable for about 90 minutes and that full access returned at 5:33 p.m. New York time (2:33 p.m. PT) on Saturday, Bloomberg reported.
What Meta did not say
Stone's update leaves the central questions open. As Bloomberg noted, the statement did not say which Claude tool was involved, who was operating it, or what it was doing in connection with the account. It also does not explain why reading activity by a tool would count against the No Kings account itself, which only makes sense if the tool was running inside a session logged in to that account.
As of Monday morning PT, we found no public comment from Anthropic on Meta's account of events.
Organizers were not satisfied. "We have regained access to our accounts, but Meta still owes us an explanation," the coalition posted on Instagram on Sunday, according to The Guardian. "Restricting communications about voting in the middle of an election is unacceptable." Organizers told the Guardian that members "posted only a handful of times and livestreamed to thousands," pushing back on Meta's first, posting-based explanation. Senator Chris Murphy called for an investigation, the Guardian reported.
Why it matters for anyone using AI agents
Set the politics aside and the technical claim is simple: reading, not posting, got the account flagged. That is the part that matters to anyone who lets an AI tool work inside their social accounts.
Anthropic ships several products that can browse on a user's behalf. Claude can fetch web pages when a user asks a question, using a user agent called Claude-User, according to Anthropic's crawler documentation. Claude Cowork has a built-in desktop browser, and Claude in Chrome runs inside the user's own browser. A tool that works inside a logged-in session reads at machine speed: it opens profiles, scrolls hashtags and pulls posts in rapid sequence. To a platform's security systems, that pattern can look the same as a scraper, whatever the intent.
Platforms built their defenses for scrapers and spam bots, not for a user's own assistant. Meta's statement describes exactly that mismatch. Its systems saw bursts of sequential requests and acted on the account. The statement gives no sign that they could tell an agent working for the account holder from an attacker. The fix took an automated re-review and then extra manual work on admin access.
For organizations that depend on one social account for a time-critical event, the practical lessons are plain:
- An AI agent reading inside your account can trigger enforcement against that account, even if you never post.
- Burst activity is the risk. Meta pointed to sharp, short spikes, not total volume.
- Keep agents out of the main account on high-stakes days, or use a separate account for research and monitoring.
- Have a backup channel ready. A suspension of about 90 minutes, plus longer admin lockout, can cost a live event its main way to reach people.
For a broader look at how agents get access to accounts in the first place, see our explainer on screen scraping, OAuth, open banking APIs and MCP.
The election context
The timing adds weight. Anthropic published its own plan for the midterms on October 14, describing election safeguards and a partnership with Democracy Works for official voting information in Claude.ai (Anthropic). Meta's account now puts a Claude tool at the center of an election-week platform dispute, even though nothing Meta said suggests Anthropic did anything beyond making a tool that someone used.
The open questions are the ones Stone did not answer: which product, run by whom, and why its reading counted against No Kings. Until Meta or Anthropic says more, the confirmed facts are Meta's own: a Claude tool read fast, Meta's systems mistook it for a bot, and the account went dark during the event.
Sources
- Andy Stone on X, October 18 update
- Andy Stone on X, October 17 statement
- Bloomberg via Yahoo Finance: Meta Says Claude AI Tool Triggered No Kings Instagram Suspension
- The Guardian: Outrage after Meta suspends No Kings Instagram account during protests
- Business Insider: Meta says it shut No Kings account because it suspected bot activity
- Anthropic: Preparing for the 2026 US midterms
- Claude Help Center: Does Anthropic crawl data from the web?
