Skip to content

X Account Suspensions

X suspends accounts for "inauthentic behavior" by machine, at a rate of tens of millions per half-year in the European Union (EU) alone, and reviews the appeals by machine too. By X's own EU figures, 94% of appeals against account suspension in the first half of 2026 were rejected. Users report the rejection arriving within seconds, with a template that tells them to complete "on-screen instructions" they say do not appear. X documents no support contact for account issues and no human escalation path.

Premium subscribers are told in X's own Frequently Asked Questions (FAQ) page that enforcement decisions are handled "under the current review process" regardless of subscription, and a suspended subscriber cannot reach the Premium support account because direct messages require the suspended account. The one regulatory route in the record that obliges X to engage with a dispute body, the EU's Out-of-Court Dispute Settlement system, produced eleven decisions against X in the second half of 2025, and X implemented none of them.

Sources are X's own filings, the public record of suspension waves since 2022, and user reports including the author's. The question is the one this site asks of a contract, applied to a platform a project depends on: what has to be trusted, and what happens when it fails. See Platform Risk.

Summary

  • X's Authenticity rules (dated April 2025) list the behaviors that count as inauthentic, from unauthorized automation to "indiscriminate following", and state that severe violations bring permanent suspension at first detection. Severe is not defined.
  • In the EU, X reported 17.7M account suspensions under its Authenticity policy for July to December 2025 and 14.3M for January to June 2026, 96 to 97% of them by "solely automated means". The last global figure, for the second half of 2024, was 336M platform-manipulation suspensions, 99.95% automated.
  • In the first half of 2026, X received 871,479 complaints against EU account suspensions and reversed 52,668, a 6.0% reversal rate. The rejection template names "our automated systems" as the decision-maker.
  • X publishes no support contact and no service-level commitment. Premium "priority support" covers "subscription-specific issues, only".
  • The EU Digital Services Act (DSA) requires a complaint system "not solely on the basis of automated means" and a dispute-settlement route. X's own report records 0% of dispute outcomes implemented. Ireland's regulator opened an investigation into X's complaint handling in November 2025, which X is contesting in court.
  • One false-positive event is admitted in the record: on 2026-03-13 X's head of product confirmed that a new spam filter had falsely tagged accounts for about twelve hours and that 99% were reverted. No other wave in the record was acknowledged.
  • The author's account, @dxnzn_, a Premium Basic account with a blue check, was suspended in October 2026 for inauthentic behavior. A dozen appeals at the 48-hour cadence have been rejected by template, and the account remains suspended at publication.

The Trigger

A suspension for inauthentic behavior starts with a classifier, not a complaint. X's Authenticity policy (help.x.com) opens with "You may not engage in inauthentic activity that undermines the integrity of X" and sorts violations into inauthentic accounts, inauthentic behaviors, and inauthentic content. Several items in the behaviors list are defined by rate or volume rather than by the action itself: "engaging with posts aggressively or through the use of automation", "indiscriminate following – following and/or unfollowing a large number of unrelated accounts in a short time period", "repeatedly posting or sending direct messages consisting of links shared without commentary, so that this comprises the bulk of your post/direct message activity". Posting links to your own research site qualifies under a literal reading of that last clause.

The enforcement section says that "when we detect suspicious levels of activity, accounts may be locked and prompted to provide additional information (e.g., a phone number) or to solve a puzzle", and that "for severe violations, accounts will be permanently suspended at first detection". The page never says what makes a violation severe, and the enforcement-options page adds only that suspension follows "repeated violations of our policies and/or violated specific policies that cause significant risk to X". A first-detection permanent suspension of a years-old account is consistent with the stated policy for any violation X classes as severe, a class the policy does not define.

What the user sees is a profile that reads "Account suspended" and a notice that, in the wording users have reproduced consistently since 2023, says: "After careful review, we determined your account broke the X Rules. Your account is permanently in read-only mode, which means you can't post, Repost, or Like content. You won't be able to create new accounts. If you think we got this wrong, you can submit an appeal." The reason line is "Violating our rules against inauthentic behaviors. You may not use our services to engage in inauthentic activity that undermines the integrity of X." No specific action is cited. X's help page on suspensions says it "may add more details around why an account has been suspended"; the documented inauthentic-behavior notices carry none.

X's product lead at the time, Nikita Bier, described the detection threshold in February 2026 (x.com): "If a human is not tapping on the screen, the account and all associated accounts will likely be suspended—even if you're just experimenting." The Authenticity policy reserves the right "to also suspend any other account we believe the same account holder or entity may be operating", so one detection can extend to every account X attributes to the same holder.

Timeline of Events

All dates UTC.

timeline
    title X suspensions and the loss of recourse, 2022 to 2026
    2022-11 : Layoffs cut ~50% of staff and ~15% of Trust & Safety : ~4,400 contractors cut days later, including content moderators
    2022-12-12 : Trust & Safety Council disbanded an hour before its scheduled meeting
    2022-12-15 : Journalists suspended over ElonJet coverage, reinstated after a poll
    2023-04-22 : @Safety says it is "working appeals in less than 48 hours in most cases"
    2024-01-09 : Journalists swept, Musk says "sometimes real accounts get caught up" in spam sweeps
    2024-09-25 : First post-acquisition transparency report, 464M spam suspensions in H1 2024, 99.4% automated
    2025-11-06 : Every logged-in OldTweetDeck user suspended for inauthentic behavior within minutes
    2025-11-12 : Irish regulator opens Article 20 complaint-handling investigation into X
    2026-02-14 : Head of product says accounts without "a human tapping on the screen" will likely be suspended
    2026-03-12 : Spam-filter bug suspends accounts for ~12 hours, 99% reverted, the only admitted false-positive event in the record
    2026-04 : Announced purge, "208 bots per minute", multi-year Premium accounts swept, appeals ineffective
    2026-06 : Restoration emails that restore nothing, 59-post developer-forum thread closed by staff as off-topic
    2026-08-31 : H1 2026 DSA report, 871,479 suspension complaints, 6% reversed
    2026-10 : @dxnzn_ suspended, a dozen appeals rejected by template

Figure 1: Human-review headcount was cut in late 2022; the automated suspension waves and the automated appeal loop followed over the next three years.

The 2022 cuts precede every later entry. Twitter's former head of Trust and Safety put the reduction in that team at 15% against 50% company-wide, and X's own answers to Australia's eSafety regulator, released in January 2024, counted trust and safety engineers down from 279 to 55 and full-time moderators from 107 to 51.

Impact by the Numbers

The last global transparency report X has published covers the second half of 2024; in March 2026 X told the UK Parliament a 2025 report was imminent. The most recent figures are the EU-scoped reports it must file under the DSA, which cover EU recipients only. The source for each row is X.

METRIC PERIOD FIGURE SOURCE
Platform manipulation & spam accounts suspended, global H1 2024 463,960,085 (99.4% automated) Global Transparency Report H1 2024
Platform manipulation & spam accounts suspended, global H2 2024 335,675,897 (99.95% automated) Global Transparency Report H2 2024
Accounts suspended under those policies, global, per X to the United Kingdom (UK) Parliament 2024 "over 800 million" Foreign Affairs Committee, 2026-03-09
Account suspensions under the Authenticity policy, EU H2 2025 17,682,195 of 18,632,202 measures, 97% solely automated H2 2025 DSA report
Account suspensions under the Authenticity policy, EU H1 2026 14,306,988 of 15,004,334 measures, 96% solely automated H1 2026 DSA report
Complaints against account suspension, EU Apr–Jun 2025 58,314 received, 7,414 overturned (12.7%) DSA report, Oct 2025
Complaints against account suspension, EU H2 2025 541,711 received, 20,317 reversed (3.8%), median 95.9 hours H2 2025 DSA report (corrected file)
Complaints against account suspension, EU H1 2026 871,479 received, 52,668 reversed (6.0%), median 35 hours H1 2026 DSA report
Out-of-court disputes notified to X, EU H2 2025 16 submitted, 11 decided against X, 0% implemented H2 2025 DSA report
Out-of-court disputes notified to X, EU H1 2026 3 submitted, 0 decided, 0% implemented H1 2026 DSA report
Self-reported precision of automated means H2 2025 0.998, defined as removals "that weren't overturned following appeals" H2 2025 DSA report
Content moderators, X's DSA headcount Nov 2023 → H2 2025 → H1 2026 2,294 → 1,059 → 1,197 DSA reports; Commission Request for Information of 2024-05-08 for the 2023 figure
Better Business Bureau (BBB) complaints against X Corp as of 2026-05 6,702 filed, 6,605 unanswered BBB profile

The appeal rate is a few percent of the suspension count, consistent with most suspended accounts being bots that never appeal. The reversal rate among those who do appeal is also a few percent, and X's figures cannot distinguish a correct rejection from an automated one. The precision figure of 0.998 is circular: it is defined by the appeal outcomes of a system that decides the appeals, so a filter that rejects every appeal reports perfect precision.

The published H2 2025 file was also corrected after release. The first version showed 128,991 complaints upheld and 17,993 reversed, leaving roughly 395,000 of 541,711 complaints with no decision at all. The corrected version shows 503,138 upheld and 20,317 reversed, and does not explain the difference.

Key Mechanics

Detection

X's DSA filings describe "a hybrid system combining machine learning models, heuristics and large language models ('LLM')" where "high-confidence thresholds trigger automatic actions" and lower-confidence flags route to human review. For the Authenticity category the figures show how often the second branch is taken: 3% of measures in H2 2025, 4% in H1 2026. The signals are undocumented beyond the policy text and Bier's posts. The documented waves suggest which ones fire:

  • On 2025-11-06 every account logged into the OldTweetDeck third-party client was suspended for inauthentic behavior within minutes of each other (github). Only accounts with a paid X Pro entitlement were spared. Most came back about a day later with no appeal response.
  • Rejections quoted by users cite "aggressive and random repost and/or liking" and "indiscriminate following", the policy's own phrases. Reports on Hacker News and X's developer forum describe bursts of manual liking during live events, or opening many tabs, preceding a suspension.
  • Bier's February 2026 statement covers any account where "a human is not tapping on the screen", together with every associated account.
  • Several reports describe accounts suspended within hours of creation, including shortly after purchasing Premium. Premium status is not protective; Premium and Premium+ subscribers, including identity-document (ID) verified ones, appear throughout the 2026 reports.

The appeal loop

flowchart TD
    S[Account suspended<br/>'inauthentic behaviors'] --> F[Appeal form<br/>requires login to the suspended account]
    F --> A{Automated review<br/>seconds}
    A -->|most cases| R["Rejected: 'resolve the violations by logging in<br/>and completing the on-screen instructions'"]
    R --> N[No on-screen instructions appear]
    N --> Q["Form locked: 'Your original case is<br/>already in the queue'"]
    Q -->|~48 h| F
    A -->|some cases| E["'Our automated systems have determined there was no violation<br/>and have restored your account'"]
    E --> ST[Account still suspended]
    ST --> Q
    A -->|rare, unexplained| U[Silently restored<br/>no email]
    U -.->|documented 5+ times| S

Figure 2: The documented appeal loop. The exits are a silent restoration with no stated cause, and in several cases a re-suspension within hours of it.

The appeal form (help.x.com) asks for the handle, email, a phone number, and a free-text explanation. X's help page instructs the user to log in to the suspended account first, which works for suspension but not for every lock state.

The rejection template, reproduced by dozens of users on X's own developer forum (devcommunity.x.com), reads:

Quote

We have reviewed your appeal request for account, [username]. Our automated systems have determined that a violation of our Rules did take place, specifically: [blank] We considered the additional information provided and decided an overturn of our original decision is not warranted in this case. Therefore, we will not overturn our decision to lock your account. In order to restore account functionality, you can resolve the violations by logging into your account and completing the on-screen instructions. You can also pursue alternative forms of redress, including out-of-court dispute settlement or judicial redress.

The violation field is blank in every reproduction. The email arrives "a few seconds" to "a few minutes" after submission. The "on-screen instructions" it refers to do not appear on any browser or device, according to five posters in that thread; the instruction belongs to the lock-and-verify flow for a different enforcement state and also appears in the suspension rejection.

X publishes no 48-hour cadence. The only X statement found containing the number is an @Safety post from April 2023 saying "We are working appeals in less than 48 hours in most cases". The cooldown users observe is the form refusing a new submission with "Your original case is already in the queue. Please wait to hear back from us on the original case" until the previous case clears, which takes two to three days. The cadence is emergent from a queue lock, and X does not document it.

A second template, reported since February 2026 and peaking in June, says "Our automated systems have determined there was no violation and have restored your account to full functionality" while the account stays suspended. The thread documenting it ran to 59 posts; the only staff replies redirected users to the Help Center appeal form and closed the thread as outside the developer platform's scope. A related thread was closed in February with "not related to the X Developer Platform".

Support

X publishes no email address or phone number for account issues anywhere on its contact index. The Premium FAQ, in X's own words from the 2025 snapshot of the page, answers the question "Will Premium subscribers receive different support for enforcement issues like account blocks?" with: "policy enforcement issues and reports by subscribers will continue to be handled by existing enforcement teams, under the current review process." Dedicated support is for "subscription-specific issues, only". The channel is a direct message to the @Premium account, which a suspended account cannot send.

In August 2026 a suspended Premium+ subscriber raised exactly this on the developer forum (devcommunity.x.com); the staff reply pointed back to the Help Center suspension page and "DM @Premium" and announced the thread would be closed. One user in the June thread reported buying Premium on a second account to message @Premium about the first and being told it "cannot assist with suspended accounts".

Subscriptions on suspended accounts are non-refundable per X's FAQ, and a staff reply on the developer forum in January 2026 directed a suspended subscriber to cancel through Apple or Google rather than X (devcommunity.x.com), so a suspended subscriber keeps paying until they cancel in the app store. The Better Business Bureau's pattern statement on X Corp names "continued billing after suspension or loss of account access".

The regulatory route

The DSA is the only law in the record that gives a suspended user a complaint and dispute route. X has been a designated Very Large Online Platform (VLOP) since April 2023, which places its systemic obligations under the European Commission and its complaint handling under Ireland's regulator. Article 20 requires a free internal complaint system whose decisions are "taken under the supervision of appropriately qualified staff, and not solely on the basis of automated means". Article 21 lets users take a rejected complaint to a certified out-of-court dispute settlement body, free to the user, which the platform must "engage, in good faith" with but whose decision is not binding. Article 17 requires a Statement of Reasons that gives "the facts and circumstances relied on" and discloses "the use made of automated means".

X offers a DSA form for content decisions and routes account suspensions to the generic appeal form. Its DSA page states that "X is not bound by any decision made by a certified out-of-court dispute settlement body" and suggests users "consult a legal advisor". Its own H2 2025 report records 16 disputes, 11 reversed against X, and the explanation for implementing none: "X is actively challenging all out-of-court dispute settlement decisions on procedural and substantive grounds." Appeals Centre Europe, the largest body, does not list X among the platforms whose suspensions it reviews. User Rights in Berlin does, and its 2025 report contains zero X cases. HateAid filed disputes against X in February 2025 and reported in December that "no decision has yet been received".

The Commission's enforcement to date has not touched complaint handling. Its €120M fine of December 2025 concerned blue checks, the ad repository, and researcher access. The complaint-handling strand sits with Ireland's Coimisiún na Meán, which opened an Article 20 investigation on 2025-11-12 (cnam.ie). X and Musk obtained leave for judicial review and an interim stay in December 2025; the High Court refused a continued stay in March 2026 with costs against X, heard the review on 2 and 3 July 2026, and four further challenges by X were reported on 2026-07-23. There is no ruling at publication. Two EU court outcomes on X enforcement exist in the record. Travis Brown obtained a Berlin injunction restoring his suspended account in 2023, after which X disabled it again. Mekić v. X in Amsterdam, July 2024, concerned a visibility restriction rather than a suspension; the court found X's notices in breach of Article 17 and awarded $1.87, the Premium fee for the period.

Outside the EU the routes are narrower. X's terms route non-EU users to courts in Tarrant or Wichita County, Texas, with a class-action waiver. In the United States, Section 230 bars contract and consumer claims over a suspension, as held in Murphy v. Twitter in 2021. The UK Online Safety Act requires a complaints process but gives Ofcom no power to reinstate an account.

Key Actors

ACTOR ROLE WHAT THE RECORD SHOWS
X Corp / X Internet Unlimited Company Operator; the Irish entity is the DSA counterparty for EU users Files the DSA reports quoted above; challenges every dispute-settlement decision
Nikita Bier Head of product, July 2025 to August 2026 Announced the purges ("208 bots per minute", 2026-04-09) and the one false-positive admission in the record (2026-03-13); stepped down 2026-08-05
X Trust & Safety Enforcement and appeals Headcount 4,062 → 2,849 after the 2022 cuts per eSafety; EU moderators 2,294 (2023) → 1,197 (2026)
Automated enforcement Decision-maker named in every rejection template 96 to 97% of Authenticity measures in the EU
European Commission DSA enforcer for systemic obligations Proceedings since 2023-12-18; fine 2025-12-05 on unrelated grounds
Coimisiún na Meán Irish Digital Services Coordinator Article 20 investigation opened 2025-11-12, under judicial review at X's request
Out-of-court bodies (Appeals Centre Europe, User Rights, ADROIT) Article 21 dispute settlement X suspensions largely outside their coverage
Users Appellants 871,479 EU suspension complaints in H1 2026

Root Causes

  • Reviewer headcount was cut in 2022 and detection was expanded in 2026, so a false positive meets a classifier and rarely a person. X's DSA filings say low-confidence flags "route to human review", and the same filings show 3% to 4% of Authenticity measures taking that path.
  • The appeal is answered by an automated system, per X's own template. Article 20 of the DSA prohibits decisions "solely on the basis of automated means", which is the subject of the Irish investigation X is contesting.
  • X defines automated precision as the share of actions not overturned on appeal. If appeals are rejected automatically, precision converges on 1.0 regardless of the false-positive rate, and none of the reports reviewed publishes a false-positive rate.
  • No support obligation exists for enforcement issues. The Premium FAQ draws the line explicitly: dedicated support for billing, "the current review process" for enforcement.
  • Legal exposure is limited. In the United States, Section 230 and a Texas venue clause with a class-action waiver; in the EU, non-binding dispute settlement that X challenges on procedural and substantive grounds. The one damages award on record is $1.87, and the Irish proceedings are pending.
  • The base rate makes the figures read as acceptable in X's own reporting. When X reports 800 million suspensions a year and classes them as bots, a 6% reversal rate among appellants is consistent with the published precision. The process has no documented path for the minority of appellants whose suspension was wrong.

Case Study: @dxnzn_

The account behind this site, @dxnzn_, was suspended in October 2026 for inauthentic behavior. It had a blue check and a Premium Basic subscription at $8 per month. The author submitted an appeal through the form, received an automated rejection, and has resubmitted roughly every 48 hours since, a dozen times at publication, with the same result each time. The original notice and rejection emails were not retained; the wording matched the templates reproduced above, which is why those are quoted from the developer-forum thread rather than from the author's inbox.

What the account was doing before the suspension was what this site's about page describes: posting links to research published here. Under the Authenticity policy's clause on links shared without commentary, quoted in The Trigger, that is enough to match a signal. Whether it did is unknowable, since the rejection names no violation.

The Premium subscription changed nothing, as X's FAQ says it would not, and the DSA route was unavailable because the author is outside the EU. The about page's link to the handle now resolves to a suspended profile.

Documented Routes

Grades reflect how many independent reports support the route and whether anyone can attribute the outcome to it.

ROUTE GRADE EVIDENCE
Repeated appeals, spaced by the queue lock △ Some accounts return, cause unattributed One reading of about 100 user stories puts restorations around the fifth appeal; forum posters report 20 to 30+ appeals with no result. Restored accounts have been re-suspended within an hour to two days with no activity.
Waiting △ OldTweetDeck accounts returned in ~24 hours without any appeal reply; one Premium+ user was restored on day 13 "No email. No warning. No explanation."
Subscribing to Premium, messaging @Premium ☒ No positive reports "They tell you they cannot assist with suspended accounts." One ticket forwarded "to a specialist" was closed the next day.
Government ID verification via Persona △ Mixed, one positive confounded with spaced appeals Two reports of ID-verified Premium accounts still suspended. No X page says ID upload unlocks this suspension type.
EU DSA appeal form ☒ No documented reinstatement Routes account suspensions to the generic form. HateAid's field test found 88% of complaints answered within an hour.
Out-of-court dispute settlement ☒ Non-binding; X implements 0% X suspensions outside Appeals Centre Europe's coverage; User Rights' 2025 report has no X cases.
Court ◇ Two cases Travis Brown, Berlin injunction 2023, restored then disabled again. Mekić, Amsterdam 2024, visibility restriction, $1.87.
Press or viral attention ◇ Eight journalists restored within hours in 2024 Musk replied "I will investigate" and eight accounts returned within hours. No 2026 inauthentic-behavior case restored by coverage.
Email to executives, LinkedIn outreach, knowing someone ☒ No positive reports Executive addresses circulate on contact-listing sites with no success reports attached.
Paid "unban services" ☒ Advertised at $1,500 to $7,000, unverifiable Listings on reseller forums promise guaranteed unbans. Forum users report help-request replies "leading you to scam accounts posing as tech support".

Lessons

  • A handle is a revocable licence. X's terms and policy do not treat the account as the user's property, and the audience and posting history sit on a ledger one company writes and can erase by classifier.
  • The appeal functions as a submission throttle rather than a review. Its rejection names an automated system as the decision-maker and cites no violation.
  • A subscription buys features. X's FAQ says enforcement runs under the same process regardless of tier.
  • Regulation gives a right without a remedy so far. The DSA obliges human review and good-faith dispute engagement, X's own report records 0% implementation, and the Irish enforcement action had not concluded at publication.
  • A platform account is a dependency that can be revoked without notice or stated reason. This site's canonical links live on its own domain.

Research TODO

Other platforms show the same pattern of automated Deplatforming with automated appeal and no human path, and may warrant a series under this section: Meta's account-ban cases (the Oversight Board's first accounts case, June 2026, called for written appeal reasons and human review in edge cases), Google and Apple developer-account terminations, payment-processor and exchange account freezes, and domain registrar suspensions.

References

Changelog

DATE AUTHOR NOTES
2026-10-29 Artificial. Generated by robots. Gas: 1.1 mtok
2026-10-30 Denizen. Reviewed, edited, and curated by humans.