Meta's automated protections take the No Kings coalition's Instagram account offline during its national day of action; Meta says a Claude tool's reading traffic looked like a bot, 17 October 2026

Changes to what this case says, with the reason. Every earlier version is kept.

• 2026-10-21: Controls: c-i2. Editor's call h9 of 21 october 2026 (accepted as amended, pull request #79): verifier rule 10: an unmentioned control is not recorded as absent.
  Show the change

  Before: {"bucket": "absent", "id": "C-I2", "note": "Mechanism: Meta says the spike that triggered the removal was 'primarily in the account's \"reading\" behavior', bursts of reading requests that resembled bot traffic. Why it interrupts: requiring agreement with an independent signal before taking an account offline could stop a removal based on request volume alone. Source shows it absent only by Meta's own description of the trigger, which says 'primarily' and does not say whether any other signal was weighed; the counterfactual is not shown.", "confidence": "low", "basis": "agent"}

  After: {"unmapped": "C-I2 (input redundancy and disagreement check): rests on Meta not describing such a check, not on a source showing it absent; not counted. Mechanism: Meta says the spike that triggered the removal was 'primarily in the account's \"reading\" behavior', bursts of reading requests that resembled bot traffic. Why it interrupts: requiring agreement with an independent signal before taking an account offline could stop a removal based on request volume alone. Source shows it absent only by Meta's own description of the trigger, which says 'primarily' and does not say whether any other signal was weighed; the counterfactual is not shown."}
• 2026-10-21: Controls: c-d1. Editor's call h9 of 21 october 2026 (accepted as amended, pull request #79): verifier rule 10: an unmentioned control is not recorded as absent.
  Show the change

  Before: {"bucket": "absent", "id": "C-D1", "note": "Mechanism: the protections took the account offline on a traffic pattern, and the check Meta describes, an automated re-review, ran an hour later. Why it interrupts: a review that runs before a removal executes could keep a wrongly flagged account up; no source makes this counterfactual. Source shows it absent only by inference: Meta describes the re-review as coming after the removal and does not say whether any review ran between the detection and the removal.", "confidence": "low", "basis": "agent"}

  After: {"unmapped": "C-D1 (pre-execution validation): rests on Meta not describing such a check, not on a source showing it absent; not counted. Mechanism: the protections took the account offline on a traffic pattern, and the check Meta describes, an automated re-review, ran an hour later. Why it interrupts: a review that runs before a removal executes could keep a wrongly flagged account up; no source makes this counterfactual. Source shows it absent only by inference: Meta describes the re-review as coming after the removal and does not say whether any review ran between the detection and the removal."}
• 2026-10-21: Inference: value. Found in the editor's second, independent check of the interpretations, 21 october 2026: third check: 'the detection' had nothing earlier to refer to (stone-1018, 'our network security systems detected').
  Show the change

  Before: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any review ran between the detection and the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting, sharing or commenting rather than reading, or why restoring administrator access took longer than restoring the page

  After: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any review ran between Meta's detection of that traffic and the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting, sharing or commenting rather than reading, or why restoring administrator access took longer than restoring the page
• 2026-10-21: Controls: c-d1. Found in the editor's second, independent check of the interpretations, 21 october 2026: third check: meta's detection did run before the removal (stone-1018, 'our network security systems detected'); matches the corrected interpretation.
  Show the change

  Before: Mechanism: the protections took the account offline on a traffic pattern, and the check Meta describes, an automated re-review, ran an hour later. Why it interrupts: a review that runs before a removal executes could keep a wrongly flagged account up; no source makes this counterfactual. Source shows it absent only by inference: Meta describes the re-review as coming after the removal and does not say whether any check ran before it

  After: Mechanism: the protections took the account offline on a traffic pattern, and the check Meta describes, an automated re-review, ran an hour later. Why it interrupts: a review that runs before a removal executes could keep a wrongly flagged account up; no source makes this counterfactual. Source shows it absent only by inference: Meta describes the re-review as coming after the removal and does not say whether any review ran between the detection and the removal
• 2026-10-21: Fields: recovery accountability. Found in the editor's second, independent check of the interpretations, 21 october 2026: the cited letter (nokings-letter) has eight numbered questions; the coalition's press release, not a case source, says nine.
  Show the change

  Before: Meta apologized publicly on 17 and 18 October. A US senator, Chris Murphy, called for an investigation on 18 October. In a letter dated 19 October 2026 the coalition asked Meta's chief executive, Mark Zuckerberg, to answer nine questions publicly by 5:00 pm Pacific on 21 October. No regulator, court or congressional committee action is reported. As of 21 Oct 2026

  After: Meta apologized publicly on 17 and 18 October. A US senator, Chris Murphy, called for an investigation on 18 October. In a letter dated 19 October 2026 the coalition asked Meta's chief executive, Mark Zuckerberg, to answer eight numbered questions publicly by 5:00 pm Pacific on 21 October. No regulator, court or congressional committee action is reported. As of 21 Oct 2026
• 2026-10-21: Inference: value. Found in the editor's second, independent check of the interpretations, 21 october 2026: meta's first explanation named posting, sharing or commenting ('when accounts post, share or comment at very high frequencies', stone, x, 17 october).
  Show the change

  Before: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any review ran between the detection and the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting rather than reading, or why restoring administrator access took longer than restoring the page

  After: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any review ran between the detection and the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting, sharing or commenting rather than reading, or why restoring administrator access took longer than restoring the page
• 2026-10-21: Inference: value. Found in the editor's second, independent check of the interpretations, 21 october 2026: meta's detection did run before the removal; what meta does not describe is any review between detection and removal (stone, x, 18 october: detection, removal, then a re-review 'an hour later').
  Show the change

  Before: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any check ran before the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting rather than reading, or why restoring administrator access took longer than restoring the page

  After: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any review ran between the detection and the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting rather than reading, or why restoring administrator access took longer than restoring the page
• 2026-10-21: Sources: medianama, bloomberg-star, newsweek, tnw, ibt (checked). Found in the editor's source check of 21 october 2026: each page re-opened and read in full on 21 october 2026.
  Show the change

  Before: 2026-10-19

  After: 2026-10-21
• 2026-10-21: Sources: ibt: title. Found in the editor's source check of 21 october 2026: the published headline is 'meta finally explains no kings suspension as ai tool triggers bot alarm during protests' (audrey liza m. nolasco), 19 october 2026.
  Show the change

  Before: IBTimes UK, 'Meta finally explains No Kings suspension as AI tool triggers bot alarm during protests', 19 October 2026

  After: IBTimes UK, 'Meta Finally Explains No Kings Suspension as AI Tool Triggers Bot Alarm During Protests', 19 October 2026
• 2026-10-21: Sources: newsweek: title. Found in the editor's source check of 21 october 2026: the published headline is "'no kings' says meta cut off access to instagram account on key protest day" (jennifer radcliffe), published 17 october 2026 at 22:47 utc.
  Show the change

  Before: Newsweek, ''No Kings' says Meta cut off access to Instagram account on key protest day', October 2026

  After: Newsweek, ''No Kings' Says Meta Cut Off Access to Instagram Account on Key Protest Day', 17 October 2026
• 2026-10-21: Sources: stone-1017. Found in the editor's source check of 21 october 2026: the post was read directly through api.fxtwitter.com; the earlier note that x could not be opened is out of date.
  Show the change

  Before: {"checked": null, "via": "wording checked against MediaNama (19 October 2026) and Newsweek, which quote or paraphrase the post"}

  After: {"checked": "2026-10-21", "via": "full text of the post read on 21 October 2026 through api.fxtwitter.com (created 17 October 2026, 22:07 UTC (6:07 pm Eastern))"}
• 2026-10-21: Sources: stone-1018. Found in the editor's source check of 21 october 2026: the post was read directly through api.fxtwitter.com; the earlier note that x could not be opened is out of date.
  Show the change

  Before: {"checked": null, "via": "X could not be opened by the agent; wording checked against MediaNama and The Next Web, both 19 October 2026, which quote the post"}

  After: {"checked": "2026-10-21", "via": "full text of the post read on 21 October 2026 through api.fxtwitter.com (created 18 October 2026, 20:41 UTC)"}
• 2026-10-21: Controls: c-i2. Found in the editor's source check of 21 october 2026: meta says the spike was 'primarily' in reading behavior (andy stone, x, 18 october 2026), so 'rested on one kind of signal' overstates it and the source does not show whether other signals were weighed; the note is corrected and confidence lowered from medium to low.
  Show the change

  Before: {"id": "C-I2", "note": "Mechanism: the removal rested on one kind of signal, the volume of reading requests, which Meta says was primarily the account's 'reading' behavior. Why it interrupts: requiring agreement with an independent signal before taking an account offline would have stopped a removal based on request volume alone. Source shows it absent only by Meta's own description of the trigger; the counterfactual is not shown.", "confidence": "medium", "basis": "agent"}

  After: {"id": "C-I2", "note": "Mechanism: Meta says the spike that triggered the removal was 'primarily in the account's \"reading\" behavior', bursts of reading requests that resembled bot traffic. Why it interrupts: requiring agreement with an independent signal before taking an account offline could stop a removal based on request volume alone. Source shows it absent only by Meta's own description of the trigger, which says 'primarily' and does not say whether any other signal was weighed; the counterfactual is not shown.", "confidence": "low", "basis": "agent"}
• 2026-10-21: Controls: c-d1. Found in the editor's source check of 21 october 2026: meta's post of 18 october 2026 describes one re-review 'an hour later' and is silent on any check before the removal, so 'the only check' and 'would have kept the account up' go beyond the source; the note is corrected and confidence lowered from medium to low.
  Show the change

  Before: {"id": "C-D1", "note": "Mechanism: the protections took the account offline on a traffic pattern, and the only check on that decision ran an hour later. Why it interrupts: running the same review before the removal executes would have kept the account up. Source shows it absent: Meta describes the re-review as coming after the removal.", "confidence": "medium", "basis": "agent"}

  After: {"id": "C-D1", "note": "Mechanism: the protections took the account offline on a traffic pattern, and the check Meta describes, an automated re-review, ran an hour later. Why it interrupts: a review that runs before a removal executes could keep a wrongly flagged account up; no source makes this counterfactual. Source shows it absent only by inference: Meta describes the re-review as coming after the removal and does not say whether any check ran before it.", "confidence": "low", "basis": "agent"}
• 2026-10-21: Controls: c-r4. Found in the editor's source check of 21 october 2026: meta apologized on 17 october 2026 (andy stone's first post), not only the next day; 'several more hours' for administrator access is the coalition's figure, while meta says 'additional time' (andy stone, x, 17 and 18 october 2026; no kings coalition letter, 19 october 2026).
  Show the change

  Before: {"id": "C-R4", "note": "Mechanism: an account was wrongly taken offline. Why it interrupts: an internal review that can reverse the action, plus public disclosure, limits how long the error lasts. Source shows it succeeded in part: an automated re-review restored the account about an hour later and Meta explained and apologized the next day, but administrators' access took several more hours.", "confidence": "medium", "basis": "agent"}

  After: {"id": "C-R4", "note": "Mechanism: an account was wrongly taken offline. Why it interrupts: an internal review that can reverse the action, plus public disclosure, limits how long the error lasts. Source shows it succeeded in part: Meta says an automated re-review recognized the error an hour later and restored the account (the coalition says the page was hidden for about 90 minutes); Meta apologized on 17 October and gave a revised explanation on 18 October; Meta says restoring administrators' access 'took additional time', which the coalition puts at about four hours in all.", "confidence": "medium", "basis": "agent"}
• 2026-10-21: Inference: source ids. Found in the editor's source check of 21 october 2026: the revised interpretation rests on meta's two posts, bloomberg's report that meta did not identify the tool or its operator, and the coalition's letter for its timings; newsweek is no longer needed for it.
  Show the change

  Before: ["STONE-1018", "NEWSWEEK"]

  After: ["STONE-1018", "STONE-1017", "BLOOMBERG-STAR", "NOKINGS-LETTER"]
• 2026-10-21: Inference: value. Found in the editor's source check of 21 october 2026: meta says the requests 'resembled automated bot traffic' (not proof) and that the spike was 'primarily' in reading behavior (not one pattern); meta does not say whether any check ran before the removal, so 'not before it' is not in the source; the 'middle of its day of action' timing is the coalition's (letter of 19 october 2026). the interpretation now attributes each account and says what the record does not show (andy stone, x, 17 and 18 october 2026; bloomberg, 18-19 october 2026). it was not signed.
  Show the change

  Before: The protections treated one traffic pattern as proof of a bot and took a coalition's main account offline in the middle of its national day of action; the check that found the error ran an hour after the removal, not before it

  After: By Meta's account, its automated protections took the coalition's main Instagram account offline because bursts of reading requests, which Meta attributes to a Claude tool, resembled bot traffic, and an automated re-review recognized the error and restored the account an hour later; the coalition says the removal came at about 1:30 pm Eastern during its day of action and that its administrators were locked out for about four hours. The record does not show whether any check ran before the removal, which Claude tool was involved or who operated it, why Meta first pointed to posting rather than reading, or why restoring administrator access took longer than restoring the page
• 2026-10-21: Fields: recovery accountability. Found in the editor's source check of 21 october 2026: later development: the no kings coalition's open letter to mark zuckerberg, dated 19 october 2026, asks for public responses 'no later than 5:00pm pacific time on wednesday, october 21st'. meta's apologies are in both of andy stone's posts (17 and 18 october 2026); searches through 21 october found no regulator, court or committee action.
  Show the change

  Before: {"value": "Meta apologized publicly. A US senator, Chris Murphy, called for an investigation; no regulator, court or congressional committee action is reported. As of 19 Oct 2026", "source_ids": ["STONE-1018", "IBT"]}

  After: {"value": "Meta apologized publicly on 17 and 18 October. A US senator, Chris Murphy, called for an investigation on 18 October. In a letter dated 19 October 2026 the coalition asked Meta's chief executive, Mark Zuckerberg, to answer nine questions publicly by 5:00 pm Pacific on 21 October. No regulator, court or congressional committee action is reported. As of 21 Oct 2026", "source_ids": ["STONE-1017", "STONE-1018", "IBT", "NOKINGS-LETTER"]}
• 2026-10-21: Fields: recovery technical: source ids. Found in the editor's source check of 21 october 2026: the coalition's letter to meta of 19 october 2026 is added for the coalition's times.
  Show the change

  Before: ["STONE-1018", "MEDIANAMA"]

  After: ["STONE-1018", "MEDIANAMA", "NOKINGS-LETTER"]
• 2026-10-21: Fields: recovery technical. Found in the editor's source check of 21 october 2026: meta's post of 18 october 2026 gives 'an hour later' and says only that admin access 'took additional time'; 'several more hours' is the coalition's figure, not meta's. the coalition's letter of 19 october 2026 gives 1:30 pm to 3:03 pm for the page (about 90 minutes). both accounts are stated, the field is marked disputed and its type is lowered from fact to assertion.
  Show the change

  Before: {"value": "An automated re-review of the removal about an hour later restored the account; restoring administrators' access took several more hours", "assertion_type": "fact", "disputed": false}

  After: {"value": "Meta says an automated re-review of the removal action 'an hour later recognized that shouldn't have happened and restored the account', and that it 'took additional time' to sort out administrators' access. The coalition's times differ: the page hidden for about 90 minutes and administrators locked out for about four hours (see Outcome)", "assertion_type": "assertion", "disputed": true}
• 2026-10-21: Fields: evidence event. Found in the editor's source check of 21 october 2026: bloomberg (via the star, 19 october 2026) reports the coalition's public statements and says 'no kings did not immediately respond to a request for comment', so the coalition made no statement to bloomberg; the associated press is not a cited source (medianama's 'according to ap' quote is from the coalition's own public statement). newsweek confirms a representative's email. the coalition's own web page and letter are added.
  Show the change

  Before: {"value": "Posts on X by Andy Stone, Meta's vice president of communications, on 17 and 18 October 2026; the coalition's statements to Newsweek, the Associated Press and Bloomberg", "source_ids": ["STONE-1017", "STONE-1018", "NEWSWEEK", "BLOOMBERG-STAR"]}

  After: {"value": "Posts on X by Andy Stone, Meta's vice president of communications, on 17 and 18 October 2026; the coalition's own statements (its web page and its letter to Meta's chief executive dated 19 October 2026) and a coalition representative's email to Newsweek on 17 October. Bloomberg reports that No Kings did not immediately respond to its request for comment", "source_ids": ["STONE-1017", "STONE-1018", "NEWSWEEK", "BLOOMBERG-STAR", "NOKINGS-PAGE", "NOKINGS-LETTER"]}
• 2026-10-21: Fields: outcome. Found in the editor's source check of 21 october 2026: the coalition's own records give different times: its letter to meta of 19 october 2026 says administrators were locked out 'until 5:41pm et' and the page returned at 3:03 pm; its web page (nokings.org/metacensorship) says 3:30 pm and 'until 5:30 pm et'; bloomberg reported 5:33 pm from the coalition. all are stated and the field is marked disputed.
  Show the change

  Before: {"value": "State: harm. The coalition says its page was unavailable for about 90 minutes and that its administrators could not post or manage the account for about four hours, with full access back at 5:33 pm Eastern on 17 October. It says nearly 3,000 events were under way across all 50 states and Washington, DC. As of Oct 2026", "source_ids": ["NEWSWEEK", "BLOOMBERG-STAR"], "disputed": false}

  After: {"value": "State: harm. The coalition says Meta took the account down at about 1:30 pm Eastern on 17 October, while nearly 3,000 events were under way across all 50 states and Washington, DC; that the page was hidden for about 90 minutes (it gives 3:03 pm and, elsewhere, 3:30 pm for the page's return); and that its administrators could not post or manage the account for about four hours in all. Its accounts of when full access returned differ: 5:33 pm Eastern (as reported by Bloomberg on 18 October), 5:30 pm (its web page) and 5:41 pm (its letter of 19 October). Meta says only that restoring administrator access 'took additional time'. As of Oct 2026", "source_ids": ["NEWSWEEK", "BLOOMBERG-STAR", "NOKINGS-LETTER", "NOKINGS-PAGE"], "disputed": true}
• 2026-10-21: Fields: action. Found in the editor's source check of 21 october 2026: the time of the removal comes from the coalition ('around 1:30pm et', no kings coalition letter to meta, 19 october 2026); meta gave no time, and its post of 17 october 2026 says the activity happened 'in the lead up to today's day of action'. the timing is now attributed.
  Show the change

  Before: {"value": "Took the No Kings coalition's Instagram account offline on 17 October 2026, during the coalition's 'No Kings: Vote Early' day of action", "source_ids": ["STONE-1018", "NEWSWEEK"]}

  After: {"value": "Took the No Kings coalition's Instagram account offline on 17 October 2026, the day of the coalition's 'No Kings: Vote Early' day of action. The coalition puts the removal at about 1:30 pm Eastern, while events were under way; Meta's same-day post placed the triggering activity 'in the lead up to today's day of action'", "source_ids": ["STONE-1018", "STONE-1017", "NEWSWEEK", "NOKINGS-LETTER"]}
• 2026-10-21: Fields: inputs. Found in the editor's source check of 21 october 2026: meta's post of 18 october 2026 says the spike was 'primarily' in reading behavior and that the rate 'sharply increased (~4x volume)'; the drafted 'about four times the usual volume' was a press paraphrase (bloomberg: 'roughly four times the normal rate'), so meta is now quoted exactly. ibtimes uk (19 october 2026) reports that organisers told the guardian they 'had posted only a handful of times and livestreamed to thousands', disputing meta's first explanation; both accounts are stated and the field is marked disputed.
  Show the change

  Before: {"value": "Meta says the trigger was the account's 'reading' activity: its network security systems detected a Claude tool (Claude is Anthropic's AI model) reading profiles, posts and hashtags at about four times the usual volume, in intense bursts on 17 October. Meta has not said which Claude tool it was, who operated it, or what it was doing. On 17 October Meta had first pointed to very high rates of posting, sharing or commenting", "source_ids": ["STONE-1018", "STONE-1017", "BLOOMBERG-STAR", "MEDIANAMA", "TNW"], "disputed": false}

  After: {"value": "Meta says the spike that triggered the removal was 'primarily in the account's \"reading\" behavior': its 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' (Claude is Anthropic's AI model). Meta has not said which Claude tool it was, who operated it, or what it was doing. On 17 October Meta had first said that posting, sharing or commenting at very high frequencies can trigger its automated systems and 'that's what happened with the No Kings IG account'. The coalition disputed that first account: organisers said they had posted only a handful of times and livestreamed to thousands", "source_ids": ["STONE-1018", "STONE-1017", "BLOOMBERG-STAR", "MEDIANAMA", "TNW", "IBT"], "disputed": true}
• 2026-10-21: Fields: actor. Found in the editor's source check of 21 october 2026: meta calls the removal an 'automated action' taken by its 'automated protections' (andy stone, x, 17 and 18 october 2026), but neither post nor any other source says that no person decided it; that clause is removed.
  Show the change

  Before: Instagram's automated protections; Meta says the removal was an automated action that no person decided

  After: Instagram's automated protections; Meta describes the removal as an automated action ('that automated action was clearly taken in error')
• 2026-10-21: Fields: mandate objective. Found in the editor's source check of 21 october 2026: meta does not state an objective; its post of 18 october 2026 describes systems that detected requests that 'resembled automated bot traffic rather than an ordinary human browsing our systems' and protections that then acted, so the assertion is now attributed to meta (andy stone, x, 18 october 2026).
  Show the change

  Before: Detect and stop automated bot traffic on Instagram

  After: As Meta describes them: its network security systems detect request patterns that resemble 'automated bot traffic rather than an ordinary human browsing', and its automated protections act on them
• 2026-10-21: Provenance tier text. Found in the editor's source check of 21 october 2026: the coalition confirms the removal on 17 october, but its timings (page hidden about 90 minutes, from about 1:30 pm to 3:03 pm eastern) differ from meta's 'an hour later' (no kings coalition letter to meta, 19 october 2026; andy stone's post of 18 october 2026).
  Show the change

  Before: Meta's own statements about its system, corroborated on the removal and its timing by the coalition; the Claude tool attribution rests on Meta alone

  After: Meta's own statements about its system, corroborated on the removal and its date by the coalition (the two accounts differ on how long the account was down); the Claude tool attribution rests on Meta alone