Mirror of https://dossier.reasoner.pl/ (Polish: https://dossier.reasoner.pl/pl/). Copy 4 September 2026, 19:1x CEST. Contact: [email protected]

● Privacy-sanitized evidence dossier · updated 4 September 2026

Paid access currently disabled

Paid Claude Max access was disabled twice in three days.

He paid Anthropic about EUR 9,200 in seven months. In the last two days of August he paid EUR 90.40 more. Anthropic then switched off his paid access twice in three days — no email, no reason, no refund. His invoices still say Paid.

For journalists. The part of this story that is new since 1 September: the customer emailed this dossier to press contacts and regulators from his Gmail address, and within hours Google disabled that account for "spam". Three identical automated refusals followed, each "reviewed by the system only"; a fourth appeal is pending. The address had been the login or recovery address for about ninety services since 2011, so one automated decision locked him out of most of his online life at once. A complaint is with the Polish Digital Services Coordinator (UKE) under the DSA, supplemented on 4 September.

Every claim on this page carries a timestamp and points to a source document. Contact: [email protected] (the Gmail address no longer receives mail). Permanent copies: Internet Archive, English · Internet Archive, Polish.

The second cutoff interrupted the finalisation of court papers intended for dispatch that same day. This was not an abstract inconvenience: time-sensitive legal work stopped mid-task. The account then showed a Free plan while Anthropic's own billing page continued to list the invoices as Paid.

A public-facing evidence hub for one Polish customer's two separate dispute tracks against Anthropic (Claude) and, since 1 September 2026, a third track against Google, which restricted the Gmail address held on file by Anthropic, the bank and the consumer authorities, and refused three appeals without stating a reason (Track C). Between 11 February and 31 August 2026 this customer paid Anthropic 617 card payments totalling approximately EUR 9,200 across company API usage and personal Claude subscriptions. Against that, he has claimed approximately EUR 4,475 plus USD 100 in twelve written submissions since 17 April: a business billing dispute stated as EUR 3,230 + USD 285 (approx. EUR 3,490 combined) for Feb–Mar 2026, and a consumer track on paid Claude plans covering a EUR 137.21 usage-metering reconciliation, repeated silent substitutions of the paid model during legitimate legal work, and — new since Aug 30 — two no-notice terminations of prepaid subscription access. As of September 1, 2026 no complaint in either track has received a substantive human resolution. The only credit or refund recorded in the entire history is USD 3.00; the bank export contains no positive card-refund entry.

Status as of 1 September 2026, 13:14 CEST (see the 4 September update below)

Paid Max access remained unavailable. The account displayed the Free plan while the Aug 3, Aug 30 and Aug 31 invoices remained marked Paid. No notice explaining either cutoff, no refund confirmation and no substantive human response had arrived.

At 13:14 CEST, because the paid workspace was still unavailable, the customer continued time-sensitive work through a separate account held by a business partner. This is stated only to document that a substitute workspace was required to keep working; no identifying details of that third party are published.

Payment and cutoff sequence — Warsaw time (CEST)

Max 20x prepaid through 3 September.

This is the last verified working point — not the cutoff time.

Paid access stopped four days before the prepaid period ended.

A Pro repurchase was required to regain access.

Upgrade to Max 5x for the period through 30 September.

Paid Claude Code access was unavailable less than 23 hours after payment.

Latest incident — paid subscriptions disabled twice without notice

Aug 30: Max 20x (EUR 180, paid through Sep 3) cut 4 days early — no notice, no refund, no explanation. On Aug 3 the customer paid EUR 180.00 for the Max 20x plan; Anthropic's own receipt #2999-2342-3004 states the service period "Aug 3 – Sep 3, 2026". The last confirmed working system read was at 18:43:01 CEST on Aug 30. The cutoff was observed later, approximately between 18:55 and shortly after 19:00 CEST — four days before the paid period ended — with the message "Your organization has disabled Claude subscription access for Claude Code"; the billing page showed a free plan. A mailbox search for the entire period shows no cancellation notice, no warning, no dunning message — nothing, before or since. The only way back into the account was to pay again: at 19:08 CEST Anthropic charged EUR 18.00 for a replacement Pro plan (receipt #2552-4707-6778). The period Aug 30 – Sep 3 was thereby paid for twice.

Aug 31 at 11:46 CEST: the customer upgraded to Max 5x — EUR 72.40 charged (EUR 90.00 minus EUR 17.60 credit for unused Pro time; receipt #2899-7850-9555, service period "Aug 31 – Sep 30, 2026").

Sep 1: paid access cut again — less than 23 hours after the EUR 72.40 payment, while finalising court papers intended for dispatch that same day. The failure was first observed after 09:35 CEST and confirmed by approximately 10:12 CEST. Claude Code returned the identical message: "Your organization has disabled Claude subscription access for Claude Code · Use an Anthropic API key instead, or ask your admin to enable access." The account holder states that this was purchased through a personal Claude account and that company details were entered for invoicing. He did not create or administer a separate organization capable of disabling the subscription. The error reproduced on every retry. The account and billing page remain reachable only as a Free plan, while the same billing page lists the Aug 3, Aug 30 and Aug 31 invoices as Paid. Free-tier web access is not the paid Max service and is not a substitute for the disabled Claude Code work surface.

Money trail (bank-verified): EUR 180.00 (Aug 3) + EUR 18.00 (Aug 30, forced) + EUR 72.40 (Aug 31) = EUR 270.40 charged within 28 days, of which EUR 90.40 was charged in the last two days of August alone — both payments made after the first cutoff, and both followed by a second cutoff. The paid access is blocked as of this update. The Revolut Business API export read on Sep 1 contains 617 completed Anthropic card-payment records: 613 debits, four zero-value records and no positive refund entry.

This is not a stand-alone incident. The Sep 1 cutoff is the newest event in the same unresolved dispute documented below: the Feb–Mar API billing claim, the complaint sequence beginning Apr 17, the Jul metering and model-substitution complaints, the Aug 26 damages claim and bank dispute, and the first no-notice subscription cutoff on Aug 30. The new event supplements that history; it does not replace or narrow any earlier claim.

What this customer paid Anthropic — the scale of the relationship

This is not a dispute with a marginal user. The figures below come from a Revolut Business API export read on 1 September 2026 and cover every card payment to the counterparty "Anthropic" on that account, whatever it funded — company API usage and personal Claude subscriptions alike. They are the customer's payments to one supplier.

Seven months, approximately EUR 9,200 paid, 617 transactions, not one refund — and not one substantive answer from a human being. The first complaint in this sequence was filed on 17 April 2026; 137 days later it is still unanswered. During that period the customer kept paying.

What is claimed, item by item

Every figure below is taken from a specific written submission, on the date shown. Nothing here is an estimate invented for this page.

Approximately EUR 4,475 claimed. USD 3.00 returned. That is 0.06%.

Across five months and twelve written submissions, the entire monetary response from Anthropic to this customer is a credit of three dollars. It arrived without explanation, was never tied to any of the numbered complaints, and no e-mail accounting for it has been found in the customer's mailbox. It also does not appear as a positive entry anywhere in the 617-record bank export — so even this single gesture cannot be reconciled against Anthropic's own billing.

Stated precisely, because it matters: the USD 3.00 is the customer's own record of a credit received. It is included here in Anthropic's favour, as the only sum ever returned. Anthropic is invited to identify what it was for.

Questions Anthropic has not answered

What was the factual basis for each cutoff on Aug 30 and Sep 1: enforcement action, automated decision, technical fault or billing fault?
Who or what is the “organization” and “admin” named in the Claude Code error on a personal subscription account where company details were supplied only for invoicing?
Why did the account show a Free plan while Anthropic's billing page still showed the Aug 3, Aug 30 and Aug 31 invoices as Paid?
Will Anthropic restore the paid Max 5x Claude Code access immediately or reverse the EUR 72.40 charge?
Why have the Apr 17 claim, the Jul 21 reconciliation request, the Aug 26 damages claim and the two supplements already sent produced no substantive human resolution?
If the contact address cannot be changed and the account settings can be reached only through that address, how does Anthropic intend to serve documents on this customer at all?

Questions Google has not answered

Which message, or which content, breached which rule? Three refusals name none.
Did a human being review any of the three appeals? Google's own support chat on 1 September said the review was "by the system only".
Why does each refusal invite "another appeal" when the outcome, the wording and the absence of reasons are identical every time?
Why do paid Google services tied to the same address continue to bill while the address itself is reported by Google's server as non-existent?

Track A — business billing dispute (EUR 3,230 + USD 285; approx. EUR 3,490 combined)

A formal billing dispute was filed on April 17, 2026 to [email protected] (support Conversation ID 215473952850015, handled by the "Fin AI Agent"). Follow-ups were sent on April 25 ("URGENT — 14 days elapsed") and again on April 25 ("FINAL ESCALATION NOTICE — Pre-Chargeback Warning"). The original complaint carried evidence attachments of ~27 MB. 137 days later there is still no substantive human resolution.

Track B — consumer plan: metering and silent model substitution

EUR 137.21 metering reconciliation, promised to a human — never delivered. On July 21, 2026 the customer filed a formal request for usage-data export (GDPR art. 15), billing reconciliation of EUR 137.21 and disclosure of the limit methodology, after the account page showed 47–49% usage consumed while an independent aggregation of Anthropic's own local session logs showed ~38%. The same day Anthropic confirmed in writing: "We're transitioning your question to one of our human support agents… we'll email you when an agent has responded." No human agent has ever responded. A written reminder of August 12 is also unanswered.

Silent substitutions of the paid model: 11 documented events between 19 and 27 August, in 8 numbered incidents. The subscription was purchased for the top model (Fable 5). Local session transcripts and an independent local runtime guard log document 11 substitution/flag events and 11 runtime-mismatch log lines — the session window is launched as --model claude-fable-5 while the runtime reports claude-opus-4-8 or claude-opus-5. Each event destroyed work in progress, and every one occurred during ordinary Polish-language legal work on the customer's own court cases.

The most telling case (incident #7, Aug 26): the platform's own auto-compaction re-evaluated the session's existing context and downgraded the model with zero new user input — while the session was drafting the damages complaint about the previous substitutions. A final guard block interrupted the very save of the incident report.

Earlier baseline from the same guard log: 12 mismatch lines on Aug 12 alone; further events on Aug 8, 9, 13 and 15. Documented substitution-related usage claimed: USD 207.76.

Incident log — every documented substitution, one row each

Times are Europe/Warsaw. Each row is reconstructed from the local session transcript (fallback record naming both models, session ID) and, where present, from an independent local runtime-guard log that compares the model requested on the command line against the model the runtime actually reports. The guard is the customer's own software; its raw lines are reproduced verbatim below the table.

The eleven guard-log lines for this period, verbatim and unedited:

Two features of this record are worth stating plainly. First, the trigger is a topic, not a request: the flagged messages are ordinary Polish legal vocabulary about the customer's own criminal and civil matters — no code, no security content, no instructions of any kind. Incidents 3 and 4 were flagged for asking about the switching itself. Second, the defect cascades: once flagged, the vocabulary stays in the conversation history, so subsequent requests are flagged again, and starting a fresh session does not help if the previous screen is pasted in (incident 2). The practical effect is that an entire class of lawful work becomes unusable on the plan bought to do it.

Anthropic's own public commitment, 11 June 2026: "Starting this week, flagged requests will visibly fall back to Opus 4.8" and "On the API, any flagged requests will return a reason for their refusal." Every incident above is silent — no visible fallback notice was given in advance, no reason was shown in the client and none was returned by the API. Anthropic also stated publicly that such interventions affect roughly 0.05% of tasks. One paying account recorded eight numbered incidents in nine days.

The two access cutoffs

Service backdrop: the customer's mailbox holds 45 statuspage notification e-mails covering 15 distinct Anthropic service incidents between Aug 12 and Aug 24, 2026 (plus further incidents Aug 31). The substitution events above are not outages — they are product behavior on a paid account.

Complaint timeline — 13 written submissions, 0 substantive human resolutions

Four promises of a human. Four case numbers. No identified human reply.

The preserved correspondence shows more than generic automation. Between 21 July and 1 September 2026, Anthropic's automated agents stated four times that the matter was being transferred to a human. Each statement carried a separate conversation ID:

The preserved record contains conflicting process messages. On 26 August Anthropic said that the request was successfully in a human queue and that the customer did not need to follow up or resend. About 29 hours later, a related User Safety thread carrying the same complaint subject was closed with: “This conversation has been closed and is no longer monitored.” The same closure text appeared again on 30 August. These were related threads, not the same conversation ID, so this page does not claim that the closure proves removal from the human queue. It records that the customer was simultaneously told to wait and told that a related channel was no longer monitored.

One automated reply, on 21 July at 17:02 UTC, engaged with the substance. It recognised the complaints of 8 and 20 July and stated that formal requests requiring account review and an official response needed direct assistance from the human support team. The customer agreed to the proposed hand-off; five hours later Anthropic issued conversation ID 215475175414438. The next customer message in that chain was a written reminder on 12 August, after three weeks without the promised response.

Three separate submissions — on 8 July, 19 July and 26 August — also received the same User Safety template word for word. It pointed to a banned-account appeal route even though the account was not banned and the submissions concerned paid-service billing, consent and damages.

No response identified as coming from a human support agent appears in the preserved chain through 1 September 2026. The first complaint in the wider sequence dates from 8 July; the first explicit promise of a human dates from 21 July. The page therefore does not collapse those two periods into an unsupported “55 days after a promise” claim.

The quoted wording, message headers and conversation IDs are retained in the original .eml files. The distinction between different conversation IDs is preserved here deliberately.

Bank and regulator status

Card dispute (Revolut Business), filed Aug 26: 15 transactions (form limit; the full scope is carried in the description and attachments). On Aug 26, 20:22 CEST the bank stated in writing that it had contacted the merchant, received no commitment, and would "review the case without the merchant's assistance." On Aug 28 one PLN transaction was rejected under card-scheme rules ("services already used"); the remaining transactions entered evidence verification the same day. A separate narrow dispute covering the EUR 137.21 of prepaid-but-undelivered usage credits is being prepared. New (Sep 1): the two no-notice subscription terminations and the EUR 270.40 charged Aug 3–31 for access now blocked are being added to the bank evidence, together with an API-verified export containing 617 Anthropic card-payment records (613 debits, four zero-value records, no positive refund entry).
ECC-Net (European Consumer Centre): case ECCPL-93131 registered Aug 26; the termination incidents are being filed as a supplement.
Pre-litigation: a payment demand with a 14-day term expires September 9, 2026. Anthropic's Consumer Terms effective 8 October 2025 identify Anthropic Ireland, Limited for EEA consumers, apply Irish law and state that EEA consumers may also file disputes in their local courts under local law. Those terms permit suspension or termination without notice for a believed material breach, legal compliance or an immediate security need; for termination of a subscription on another basis, they state at least 30 days' written notice and a pro-rata refund. Anthropic has not identified which basis, if any, it relies on here. The invoices in this record identify Anthropic, PBC and include company/VAT billing details. The customer therefore reserves — rather than presents as finally determined — consumer status, the contracting entity, governing law and forum pending assessment of the facts and mandatory law.

Evidence preserved

This page publishes a privacy-sanitized summary. The underlying material is preserved in original form and can be supplied to regulators, the bank, courts and journalists for verification.

Original correspondence

The complete outbound complaint chain and Anthropic responses in original .eml form with headers, recipients, dates and message identifiers.

Payments and plan records

Anthropic receipts and invoices for Max 20x, Pro and Max 5x; subscription confirmations; the bank export and transaction statements.

Cutoff evidence

Contemporaneous screenshots of the Claude Code error, the Free-plan account state and paid invoices, preserved with SHA-256 checksums.

Google record

The restriction page, the three appeal texts and three refusals with headers, the Google One support transcript, the seven bounce notices, the 4 September server rejection, and the state e-Delivery proofs of dispatch and receipt for the complaint to the Coordinator.

Work interruption evidence

Session records, local runtime-guard logs and contemporaneous work logs documenting model substitutions and interrupted legal work.

Evidence boundary — what this page does and does not claim

The metering discrepancy (47–49% vs ~38%) is the customer's independent aggregation of Anthropic's own local session logs against the account page. Anthropic's internal metering records are required for a full join; that is exactly what the Jul 21 request asked for and never received.
Substitution events are documented from local session transcripts (fallback records with model identifiers, timestamp offsets and session IDs) and an independent local guard log. Event times on this page are normalized to Europe/Warsaw. This page does not claim intent — it documents behavior, frequency and impact on paid work.
The Aug 30 / Sep 1 disabling events are documented from Anthropic's own receipts and confirmation e-mails, the verbatim on-screen error message (screenshots preserved with SHA-256 hashes), and the bank account's API records. The customer makes no allegation as to motive; the sequence — formal claim Aug 26, bank dispute opened late Aug, first disabling Aug 30, second disabling less than 23 hours after a new payment Sep 1 — is stated as fact, and Anthropic has been asked in writing to explain it.
The statuspage archive documents service instability in the same period; it is listed as context, not as part of the substitution claims.
All amounts are taken from written complaints and bank records. No private identifiers, addresses or payment data are published. Full documents (e-mail originals with headers, transcripts, logs, bank statements) are available to journalists and regulators on request.

Reporter-ready summary

Press contact: [email protected]. Journalists may use this address to request the preserved source records and a factual response to follow-up questions.

Scale of the relationship: 617 card payments to Anthropic between 11 Feb and 31 Aug 2026 — PLN 28,066.75 + EUR 2,599.15 + USD 18.52, approximately EUR 9,200, of which roughly EUR 6,800 in the three months March–May. Bank-verified via API export; no positive refund entry in the whole file.
Amounts claimed: approx. EUR 4,475 + USD 100 across both tracks, itemised by submission and date. Received to date: USD 3.00 — about 0.06%.
12 written submissions from Apr 17 to Sep 1, 2026, including follow-ups, reminders and three supplements. Substantive human resolutions: 0. Four separate automated messages promised a human and supplied four conversation IDs; no reply identified as human appears in the preserved chain through Sep 1. Day 137 and counting.
11 documented silent substitutions of the paid model between Aug 19 and Aug 27, in 8 numbered incidents, each listed individually with time, session identifier and guard-log line — including one triggered by the platform itself with zero new user input, while the session was drafting the complaint about the substitutions.
Prepaid Max 20x (EUR 180, paid through Sep 3) disabled Aug 30, four days early, without notice; access restored only after a EUR 18 repurchase; upgraded to Max 5x for EUR 72.40 on Aug 31 — and paid Claude Code access disabled again less than 23 hours later, while finalising court papers for same-day dispatch. The account then showed Free plan while three invoices remained marked Paid.
Bank confirms the merchant did not respond to the card issuer either (Aug 26).
Separate user reports and one press-reported account-enforcement case are listed below as context only. They are not evidence of this customer's claim and do not establish a common cause.

Comparable public reports — allegations by other users, not proof of this case

Claude Code issue #5088 reports a Claude account disabled after payment for Claude Code.
A Google Play support thread records a user report that a Claude Pro subscription became unusable shortly after purchase.
COOLak's public GitHub billing dossier and its reporter brief document a separate Anthropic billing dispute. No press pickup of that dossier is asserted here without a verifiable citation.
Broader account-enforcement controversies have reached the press, including TechCrunch's report on Anthropic temporarily banning OpenClaw's creator. That case is context, not an allegation that its facts match this billing dispute.

External links reviewed on 1 September 2026. User-authored reports are presented as allegations; the TechCrunch item is secondary reporting about a different account-enforcement event.

Track C — Google: the address on file was switched off, and three appeals were refused without a reason

This track began on 1 September 2026, when the customer used his Gmail account to send the complaint supplement and press notices about the Anthropic dispute (see the postscript). Later that day Google restricted Gmail on the account "for unwanted content". Everything below is taken from Google's own messages, Google's support chat transcript and the customer's mail store.

Complaint supplement and press notices sent from Gmail. Seven addresses bounced; several recipients replied.

Gmail restricted "for unwanted content". No message, recipient or rule identified. Other Google services remained available.

Through the form on the restriction page. The same evening, in Google One support case 1-7013000041523, two agents wrote that the appeal is "reviewed by the system only" and that there is "no direct team" to contact.

"We reviewed your appeal and decided to keep restricting access." No reason. The message invites "another appeal".

Confirmation at 12:24; refusal at 17:09 in identical wording. The notice referred to a "verifier" without saying whether a person was involved.

Refused after five hours, again in identical wording, again inviting a further appeal.

Cites the complaint to the Polish Digital Services Coordinator filed at 16:45. Google confirmed receipt at the contact address: a "verifier" will review it, "most requests take 2 business days".

A test message from another mailbox was rejected by Google's server within three seconds: status 5.1.1, "The email account that you tried to reach does not exist".

What the restriction took with it. The Gmail address was the login and the contact address of the Anthropic account in dispute, the address on file with the European Consumer Centre, the bank dispute and several public authorities, and the address on which Google itself sends the receipts for the customer's paid Google services: a Google AI Pro subscription that continues to renew and a Google Cloud account with roughly EUR 5,100 of payments on record. Since 1 September none of those senders can reach the customer at the address they hold, and anything they send is not delayed but rejected.

What Google has said, in full. Three refusals with the same three sentences: the appeal was reviewed, access stays restricted, the account "was used in a way that violates Google's policies". No message, recipient, date or rule is identified in any of them. Each ends with the sentence that EU residents "may have additional options for resolving the problem", which is Google's own pointer to the Digital Services Act procedures.

What the customer has done about it. On 4 September 2026 at 16:45 CEST a complaint against Google Ireland Limited was delivered to the Polish Digital Services Coordinator (the President of the Office of Electronic Communications, UKE) under Article 53 of Regulation (EU) 2022/2065. It was sent through the state e-Delivery system with a proof of dispatch and a proof of receipt timed 16:45:04, signed with the national trusted profile, and carries thirteen annexes: the appeal texts, all three refusals, the support chat transcript, the restriction page, the list of services tied to the address, the seven bounce notices, Google's published sending limits and Gmail policies, the 4 September rejection notice and the record of paid Google services. The complaint asks the Coordinator to examine whether Google complied with the statement-of-reasons duty (Article 17), the internal complaint-handling duty including non-automated review (Article 20(6)), the diligence duty towards a paying customer (Article 14) and the single point of contact duty (Article 12). The third refusal, which arrived one minute after the complaint was delivered, will be filed as a supplement.

The data is not retrievable either. A full Google Takeout export of the account, covering 62 services, was requested on 1 September. Google's own status line for it reads "Could not export". Until a working export exists, the account holder has no copy of the mailbox that Google switched off.

One address, dozens of accounts. The restricted address had been the account holder's main identity on the internet since at least April 2011. A local copy of the mailbox, secured on 3 September, holds 9,674 messages to that address from 159 sender domains; 90 of those domains sent account correspondence: registration confirmations, verification codes, sign-in alerts, password resets, receipts and invoices. Among them are the registrar of the customer's internet domains, payment providers and more than a dozen paid subscriptions. Several services were only ever accessed with "Sign in with Google", so they have no password of their own; Google's own security alerts in the mailbox list them. Because Google's server now rejects mail to the address as non-existent, none of these accounts can be moved to another address: every provider sends its confirmation to the old one, and a password-less account cannot be entered once its session expires. The restriction is formally limited to one service. In effect it switched off an identity used across dozens of them. A first supplement to the Coordinator complaint, with the list of affected services, was prepared on 4 September. Article 54 of the Regulation gives a recipient of the service the right to compensation for damage caused by a provider's breach of its obligations. The account holder reserves that right, including proceedings before a Polish court, and is preparing the claim file.

This page records the chronology and Google's own words. It does not allege that Google acted at Anthropic's request, that the two companies coordinated, or that the restriction was retaliation for the press notices. Google's stated basis is reported only as it appeared to the account holder.

Postscript — what happened to the mailbox used for the press correspondence

On 1 September 2026, between 15:49 and 15:53 CEST, the customer used Gmail to send the complaint supplement and four press notices about this documented dispute. The six transmissions covered 107 unique addresses: published editorial contacts, consumer and public authorities, and two private copies. Seven addresses generated visible hard-failure notices; the available mailbox copy contains replies or acknowledgements from several recipients.

Later that day Google disabled Gmail on the sending account for alleged unwanted content. The account holder filed appeal restriction ID 54. In Google One support case 1-7013000041523, two support representatives stated that the appeal was system-reviewed, that Google One could not see its status, and that documents attached to the support case would not reach the appeal system.

This note records chronology, not coordination or motive. It does not allege that Anthropic asked Google to act, that Google acted in retaliation, or that the two companies coordinated. Google's stated basis is reported only as it appeared to the account holder. The separate Google record, including the appeal and support transcript, is preserved independently.

4 September 2026 — what was sent, and how they answered

Sent today. Four documents left the customer's working mailbox and the state e-Delivery box on 4 September: at 16:07 a formal notice to Anthropic with nine annexes; at 16:19 and 16:25 two follow-ups to Anthropic with the company's court-register extract, a correction of one identification number and the account owner's written authorisation; at 16:45 the complaint against Google to the Polish Digital Services Coordinator, with thirteen annexes and a state proof of delivery (see Track C).

Google. The first and second appeals were refused on 3 September, at 10:24 and 17:09 CEST; the third, filed at 11:39 on 4 September, was refused at 16:46, one minute after the complaint reached the Coordinator, in the same wording as the two before it. At 14:45 a test message to the restricted address had been rejected by Google's server within seconds: "The email account that you tried to reach does not exist". Mail sent to the address on file with Anthropic is therefore not delayed; it is not delivered at all. At 17:28 the customer filed a fourth appeal, citing the Coordinator complaint by reference; Google acknowledged it at the working contact address within the same minute. At 17:47 a new Google Takeout export of 27 services was started; Google's confirmation also reached the contact address.

Anthropic. At 16:07 CEST the customer sent Anthropic a formal notice of change of address for service, with nine annexes, from a working address. Anthropic replied five times within 19 minutes, all automated; the three Fin replies were explicitly marked as composed by an AI agent:

16:07 — User Safety: a template pointing to the help centre and the appeals form.
16:09 — Fin AI Agent: "Our system does not support changing the contact email separately from the login email." The customer was advised to keep replying from the new address "while CC-ing" the address that no longer exists.
16:20 — after the company's court-register extract was supplied: the account owner "must authorize this change"; the documentation "supports your request".
16:26 — after the owner's written authorisation: "As an AI agent, I cannot directly update billing contact information in our system or place documentation on file. To update billing details, you'll need to make changes through your account settings." Also: "The system does not generate case reference numbers for support conversations."
16:26 — User Safety, same minute: "This conversation has been closed and is no longer monitored."

Account settings can be reached only through a sign-in link e-mailed to the address on file, which is the disabled address. The route Anthropic pointed to is closed by the problem the notice describes. Between 16:20 and 16:26 the same agent first wrote that the customer's documentation "supports your request" and that the owner's authorisation was what was needed, then, once the authorisation arrived, that it "cannot directly update billing contact information" and that "the system does not generate case reference numbers". The requests for confirmation by a named person and for a position on the disputed sums remain unanswered. Since 17 April the customer has written to Anthropic sixteen times; every reply on file is automated except one message of 29 May asking whether the problem persisted.

Regulators. ECC-Net registered a second file, ECCPL-93587 (change of contact address), on 1 September. At 18:41 on 4 September a first supplement to the Coordinator complaint was delivered by state e-Delivery, with Google's third refusal, the fourth appeal, an inventory of the services tied to the disabled address and Dropbox's own documentation of its email-change procedure; the state receipt shows delivery within two seconds.

The lock is complete. At 19:03 the customer tried to add his working address to his LinkedIn account. LinkedIn first sends a six-digit code to the current primary address, which is the Gmail address Google disabled. Dropbox requires a confirmation at the same dead address before it will change it. Cloudflare, which fronts this very page, sends its password and email changes there too. The customer is signed in to these services today only because the sessions were open before 1 September; he cannot change the address anywhere, and cannot recover access anywhere once a session ends. One automated decision by one company has locked him out of his own online identity, and every other provider's security procedure now enforces that decision. He is not able to fix this himself. On 4 September the customer decided to bring a civil action before a Polish court against Google for the disabling of the account and its consequences (Article 54 of the Digital Services Act; national law on damages); the statement of claim is being prepared.

All times CEST. The full exchange is preserved in the customer's mail store.

Right of reply and correction policy

Anthropic is invited to provide a substantive response addressing the documented charges, plan status, access cutoffs and unresolved complaint history. Any factual correction supported by primary records will be reviewed promptly and, if substantiated, reflected on this page with a dated correction note.

Published 1 September 2026, updated 4 September 2026; last status check at 19:10 CEST, Warsaw (Europe/Warsaw). All displayed event times use Warsaw time. This is a consumer-interest publication of documented facts about a commercial dispute. The page distinguishes the customer's records from third-party allegations and does not allege motive for the access cutoffs. Press contact: [email protected].