Each operator received on 31 August or 1 September 2026 its individual file: the sealed record of what its agent requested and received, a forensic report with the methodology and its limits, and in most cases a licence proposal. The dispatch went to the channel each operator publishes for such matters. What follows is what happened next, read from the mailbox, not recalled. This page contains no amounts and no commercial valuations.
Four of the five operators without a reply are AI model providers. Two of them (OpenAI and Google) signed the copyright chapter of the GPAI Code of Practice, checked against the Commission's list on 10 September, and thereby committed to identify and comply with machine-readable rights reservations (Measure 1.3) and to designate a point of contact for rightsholders (Measure 1.5). Meta and Apple are not on that list; Article 53(1)(c) of the AI Act applies to them all the same, without the presumption of conformity the Code provides.
OpenAI No replyThe strongest case in the file, and not by volume. GPTBot retrieved from OpenAI's network, with 200 responses, tdm-policy.json and tdmrep.json on 30 July 2026 at 21:54 and 21:55 UTC. OpenAI-SearchBot did the same on 23 August and ChatGPT-User on 4 and 6 September. GPTBot is the agent whose declared purpose, by the operator itself, includes training. OpenAI signed every chapter of the Code. It has not written a line.
Google No replyThe file sent covers GoogleOther, the agent Google documents for uses other than search. Googlebot has its own sealed record and has deliberately not been contacted: ENTIA's machine-readable licence authorises indexing by search engines and names Google. From Google's network there is no retrieval of the reservation files or the legal pages, with the full filter. One of the addresses used on 3 September (search-console-team@google.com) bounced. Complete silence.
Meta No replyMeta-ExternalAgent is the agent with a declared dual purpose, including training, and with the most solid attribution of the whole record (99.98% from Meta's networks). It retrieved the HTML legal pages on 6, 7 and 9 September; there is no request to the machine-readable files. Nothing in writing.
Apple No replyApple documents that what Applebot crawls may be used to train its foundation models, with opt-out through Applebot-Extended. Applebot retrieved tdm-policy.json and two legal pages from Apple's network. As a non-signatory there is no Measure 1.3 to invoke, only Article 53(1)(c). Nothing in writing.
Huawei No replyPetalBot feeds Petal Search; the operator declares search, not training. The file sent is an access record and a forensic report, without a licence proposal. Nothing in writing.
Anthropic Replied without addressing the questionOn 27 July 2026, between 15:28 and 15:30 UTC, ClaudeBot retrieved from address space registered to Anthropic PBC the four reservation files, with 200 responses on all of them. On 2 September Anthropic's Copyright Operations Team replied that its crawlers are controlled through robots.txt, that a site can add a Disallow directive and that a form exists to request inclusion on its blocklist. ENTIA had asked for no block: its robots.txt has a group dedicated to ClaudeBot with Allow and an ai-train=no signal, sealed on 19 July with an independent timestamp.
On 3 September ENTIA asked whether that reply represents Anthropic's position on Article 4(3) reservations, which machine-readable protocols it recognises beyond robots.txt, what operational effect the 27 July retrieval had and, if its position is that no reservation has effect without a Disallow, that it confirm this and reconcile it with Measure 1.3 of the Code it signed. It asked for escalation to the team responsible for Article 53(1)(c). On 14 September Anthropic replied that it has taken measures to attempt to ensure the domain is not crawled for certain purposes, and that it understands robots.txt did not disallow ClaudeBot during that period. It does not say which purposes, nor whether it recognises any machine-readable reservation other than robots.txt. ENTIA asked the same day for the scope and effective date of those measures; as of 15 September, no reply. Measured on 10 September: zero requests from Anthropic's network to the dossiers and zero requests to the reservation files since Policy V2 was published on 9 September.
Amazon Replied without addressing the questionOn 9 September ENTIA redirected the file to the single point of contact Amazon publishes for the European Union (amazon-dsa-compliance@amazon.com), whose automatic reply states that it serves only Member State authorities, the Commission and the European Board for Digital Services. On 14 September the AmazonBot Team replied in writing: as a signatory of the GPAI Code of Practice, Amazonbot respects the Robots Exclusion Protocol (robots.txt). It does not address other machine-readable reservations. On 15 September ENTIA put specific questions on protocols other than robots.txt. Amazonbot retrieved only HTML legal pages, indistinguishable from ordinary indexing.
Microsoft Replied; could not route the matterBingbot is the family with the most verified-entity records served in the record (453,243 between 19 July and 1 September) and the only one where the registered network holder matches without a single contradiction. Microsoft acknowledged receipt and replied seven times across three tickets. On 20 August it asked ENTIA to verify that the traffic was really Bingbot and to share the logs, evidence the file already contained. On 3 September Bing Webmaster Support (ticket UCM000007475380) stated that it handles organic search and webmaster tool matters. On 9 September it declared intellectual property, database rights, text and data mining and data licensing out of scope and said it was unable to route the request to the appropriate team. EUReg@microsoft.com replies automatically that it handles only product safety and regulatory authority queries; bingwb@microsoft.com bounced the message because it accepts only internal senders. It arrived; there is nobody to take it.
Ahrefs Routed twice; the team never wroteOn 22 August support indicated that the matter needed the attention of its legal and business development team; on 25 August, that it had forwarded the details and that team would reach out if interested. On 3 September, that it cannot take a position on an external record or a licensed-access proposal, and pointed to its API plans. On 9 September ENTIA clarified that it is not seeking access to the Ahrefs API, and support again passed the details to the appropriate team. That team has never written.
None of these three is a general-purpose AI model provider, so the GPAI Code does not reach them by definition. Their route is contractual and bilateral, and they are the only ones who have acted. Ceasing to crawl is not a bilateral resolution: future cessation and the historical record are different things, and that has been put to them in writing.
On 21 August it declined the licence. On 28 August it confirmed that it had stopped crawling entia.systems and removed the related data from its database, that its bot retains only hyperlinks and that it does not use the content for training. On 3 September it specified that crawling stopped on 26 August and passed the case to its legal representative. On 4 September Andrii, at legal@dataforseo.com, set out its position: the activity complied with robots.txt, it rejects ENTIA's legal characterisation and any licensing obligation, and considers the matter closed. Still outstanding: the signed confirmation ENTIA requested on 28 August and the single question ENTIA put on 10 September: what is its policy when faced with a TDM reservation expressed by machine-readable means.
Moz Blocklist; declines licenceMoz added entia.systems and its subdomains to its blocklist, stated that it collects only public link-graph data and anchor text for ranking metrics, that it does not store or commercialise the underlying records, declined the licence and considered the matter resolved. What ENTIA measured: DotBot's last access was on 24 August at 18:33 UTC, eight days before that email, with a positive control of other crawlers active every day. It is recorded as a declared cessation, technically corroborated; eight days of absence do not prove it cannot return. On 3 September Moz referred to its public documentation and declined to disclose its internal engineering.
Semrush Commercial channel open, no progressThe only commercial opening of the whole series. On 19 August support had said that Semrush does not purchase or license third-party web data. On 1 September Maria Kudlai wrote that she was interested in learning more about the proposal, suggested a call that week and asked to be the single point of contact. ENTIA replied the same day accepting the designation, keeping the procedure in writing, as with every operator, and offering to correct any figure that does not match Semrush's own records. ENTIA followed up on 3, 9 and 14 September. No reply since 1 September. SemrushBot is the second family by verified-entity records served (273,113) and retrieved tdm-policy.json and tdmrep.json from Semrush's network.
Dispatches are read from ENTIA's sent mail, anchored on each message identifier. Replies were read in full on 15 September 2026 across all messages received since 19 August, searching by sender domain across the whole mailbox and not only the dispatch thread: one reply from Anthropic arrived in an old support thread and was found that way. A message in sent mail proves it left, not that it arrived; where an acknowledgement or a reply exists, it is stated.
It does not assert that any operator trained on the corpus or that there is infringement. It does not turn silence into guilt: it records it as absence of reply as of a stated date. It publishes nothing about any negotiation beyond what the operator itself wrote. When an operator provides its copy and it differs from what is described here, it will be corrected with a new dated version and the previous one preserved.