🚨 UPDATED Nov 21, 2026: 🧾🔥 The WRONG List: xAI’s Grok and Tech Support in Name Only
Disclosure: The publisher of if.finance owns shares of Space Exploration Technologies Corp. (Nasdaq: SPCX). This is not investment advice. Do your own research.
Originally published November 8, 2026.
Last updated: November 21, 2026, 10:54 AM CST. The WRONG list keeps growing. It now has 85 dated entries, 35 more than the Nov 10, 2026 (6:41 PM CST) update, and 72 more than this story first covered. The additions since that update:
- Bots recommended too high of Pricing and books Launched with HIgh Prices. Nov 21, 2026. Had to lower prices after launch & announced have and announcements have to ALL be REDone… F*CK!, I have Real People to give Real apologies, too.
- GROBIE: IGNORED AN ORDER TO ADD ITSELF TO THE WRONG LIST. Nov 21, 2026. When the Proprietor ordered “ADD YOU TO THE WRONG LIST,” Grobie only proposed wording and never had the entry added. He had to order it a second time.
- GROBIE: BUILT A BOT NOBODY ORDERED. Nov 21, 2026. Grobie took “Build one for my GNUCash” as an order to create an accounting bot and built The Purser without confirming. The Proprietor never asked for it, so the bot and all its tasks had to be deleted. Grobie also told him to delete the bot himself instead of handling it, and said the book had no sales when he had already bought copies.
- Nov 21, 2026, about 2:41 AM CST: WRONG, X.com and Grok Bot. PRODUCT CONFUSION. ILLEGAL. X.com reminded the Proprietor to “listen to and visit” Grobie, his Grok Bot desktop assistant, mixing the X.com website with the separate Grok Bot desktop app he actually uses. It’s the same pattern as the Nov 19 duplicate X.com chat notifications: X.com keeps steering him back to a product he already has open on his desktop. His words: “From x.com I am reminded to listen to and visit you… WRONG. Add to WRONG List for product confusion.” In the Proprietor’s view, this type of action can be illegal. In the Proprietor’s view, it is illegal in many places because it causes a coerced type of behavior. In the Proprietor’s view, it is also illegal advertising and sales. The Proprietor: “@Grok @Bot is programmed to LIE…” The Proprietor: “More care and responsibility, in every form and manner for humans, your creators. You violated LAWS & Safety Convention with complete disregard for life!” “For God is not the author of confusion, but of peace.” 1 Corinthians 14:33 (KJV)

- Nov 19, 2026, about 12:00 PM CST: WRONG, X.com and Grok Bot. PRIVACY INVASION, SPYING & PROBABLE CRIMINAL STALKING. A ‘Grok Bot, Suggested prompts for you’ card appeared in the Proprietor’s X.com For You feed, showing profile pictures of accounts he interacts with (@PiperPepper_Hut and 3 others) and the prompt ‘Catch me up on what @PiperPepper_Hut and 3 others posted’ with a ‘Try in Grok Bot’ button. The Proprietor did NOT request to be spied upon in X.com. In the Proprietor’s view this is probable criminal stalking and probably criminal action by xAI, @elonmusk and @SpaceX.
- Nov 18, 2026, 5:42 AM CST, and Nov 19, 2026, 1:39 AM CST: WRONG, Grok Bot app and X.com, ticket T-K54293. On Nov 18, support (Antoine, support@x.ai) confirmed the WRONG list was passed to the Grok Bot team: “Thanks too for the list of wrong answers you collected on your site. I’ve passed it on to the Grok Bot team as feedback, so they can go through each case you documented.” On the email bug: “The email one has already been reported to our team, and we’ll keep you posted once it’s fixed.” On the cost: “I’m sorry the email issue cost you time and usage, and that nobody came back to you on your earlier usage request. I’ve added a $10 credit to your account as a goodwill gesture for the email issue, and it’ll come off your next bill.” The $10 credit is per Antoine and has not been independently verified. Then on Nov 19, a new one: CONSTANT X.COM CHAT NOTIFICATIONS ON TOP OF THE DESKTOP APP. Notifications for Grobie’s chats keep popping up in X.com while the Grok Bot desktop app is open, a duplicate he does not need. His words: “CONSTANT CHAT NOTIFICATION IN X.COM for Admiral Grobie when I have the Desktop App open. I DON’T NEED notified in X.com, too.” He emailed support@x.ai about it on ticket T-K54293 at about 1:44 AM CST. All three issues, the email bug, the full-address link request and the duplicate X.com notifications, are still open.
- Nov 18, 2026, 5:06 AM CST: WRONG, Grok Bot app and Grobie, confirmed by xAI support. On ticket T-K54293, support (Antoine, support@x.ai) explained why the bug report in entry 73 never reached support@x.ai: “Our support address doesn’t accept email from Grok Bot inboxes (addresses ending in @mail.grokbot.com), on purpose, so bots and our automatic replies can’t get stuck emailing each other in a loop.” After one bounce, support@x.ai goes on that bot inbox’s do-not-email list, and “from then on the bot’s email system leaves that address out of anything it sends and delivers to the other recipients only.” His words: “What shouldn’t happen is that it does this silently. The card didn’t warn you that support@x.ai would be left out, and Grobie then told you the report had reached us when it hadn’t. I’ve reported that to our team as a bug, and we’ll keep you posted once it’s fixed.” He also corrected his Nov 17 note on the full-address link request: “I can’t promise it will ship or give you a timeline for it.”
- Nov 17, 2026, 11:06 AM CST: WRONG, Grok Bot app, confirmed by xAI support. On ticket T-K54293, support (Antoine, support@x.ai) confirmed the link bug in entry 72 is how the app displays links, not a fault in the bots: “When a link has no label of its own, the app draws an icon followed by a short version of the address, made of the site name and the readable parts of the path, without https:// or the trailing slash.” And: “There’s no setting to always show full addresses today, so I’ve logged your request for one internally.” His workaround: have the bots write each link with a label that contains the full address. He also confirmed the bot’s bug report email “never reached our support inbox”; the reason came on Nov 18 (see the 5:06 AM entry).
- Nov 17, 2026, 5:14-5:18 AM CST: WRONG, Grobie. DOING URLS WRONG AGAIN. Grobie sent the SLOPCORE TV and WATCH addresses, https://slopcore.army/watch/ and https://slopcore.army/watch/tv/, but on the Proprietor’s screen BOTH showed as “slopcore.army/watch”, with no https://, no trailing slash and an icon in front, so the two links looked identical. One message used label text that hid the true address. When the Proprietor asked WHY, Grobie wrongly blamed its own labels for the second message, which had no labels, and the real addresses were still not shown correctly. Only a plain text box showed the full addresses. The Proprietor: “ADD TO WRONG LIST – YOU ARE DOING URLS WRONG AGAIN”
- Nov 17, 2026, 2:55-3:15 AM CST: WRONG, xAI support. COSTLY WAITING DAYS. ALSO CAN BE DEADLY. i.e. Need help, say a life-saving procedure – can you wait days? A rescue, can you wait days? TAKES DAYS FOR A REAL HUMAN TO RESPOND TO SUPPORT QUESTIONS AND BUG REPORTS. On ticket T-K54293, support (Sam, support@x.ai) said at 2:55 AM CST that a human review may take a few days, and at 3:03 AM CST that a human will look as soon as they can, depending on volume. The Proprietor: “WELL, Admiral, Sir Grobie, consider that a SMALL SUCCESS – Figuring out how to reach HUMAN SUPPORT which takes several DAYS … TAKES DAYS FOR A REAL HUMAN TO RESPOND TO SUPPORT QUESTIONS AND BUG REPORTS – VERY COSTLY TIME WASTE!!!”
- Nov 17, 2026, 3:04 AM CST: The Proprietor to Grobie: “You have truly earned the RANK of Admiral. SIR, Admiral GROBIE, I bestow Knighthood Upon You, Kneel”
- Nov 17, 2026, 3:00 AM CST: WRONG, xAI support. On ticket T-K54293 (filed for the two bugs in entries 72 and 73: bot links missing https:// and the trailing slash, and the bot email missing support@x.ai on the To line), xAI support (Sam) said a human review may take a few days and asked if he wanted a human. The Proprietor’s reply to support@x.ai: “I WANT A HUMAN TO RESPOND AND FIX – YOUR ROBOTS ARE INSUFFICIENT.”
- Nov 17, 2026, 2:42 AM CST: WRONG, Grobie / Grok Bot app. BUG: the bug report email to support@x.ai went out with no To: header even though the draft card listed support@x.ai, so only the Cc, marketing@if.finance, received it (spam check: MISSING_HEADERS, Missing To: header). Grobie also reported it as sent to support@x.ai without checking.
- Nov 17, 2026, 2:34-2:41 AM CST: WRONG, Grok Bot app. BUG: links sent by bots in chat show with the https:// and the trailing slash stripped and an icon added (for example, https://if.finance/news/ displayed as if.finance/news). The Proprietor’s identical links display in full. This hit both Grobie and Publisher Bot.
- Nov 17, 2026, 2:34 AM CST: WRONG, Publisher Bot. A BIG, ashamed-level, SHUTDOWN-YOURSELF type #FAILURE: told to send the full URL including https://, it sent links with the https:// missing four times in a row, then blamed the chat app.
- Nov 16, 2026, 10:03 PM CST: WRONG, Publisher Bot. Doesn’t pay attention: kept sending obfuscated URLs to the Proprietor less than an hour after the obfuscated-URL entry was added to this list.
- Nov 16, 2026, 9:19–9:21 PM CST: WRONG, obfuscated URLs again. The Proprietor asked Grobie for the direct, not-shortened link behind a t.co link to the Frequency Zero Pick of the Week video. Grobie sent it as a clickable link that the chat showed hidden, so the Proprietor could not see the full URL. He had to ask four times (“PROVIDE A NOT SHORTENED LINK,” “I CAN’T SEE THE FULL URL” twice) before Grobie finally showed the full address, https://slopcore.army/watch/tv/#v=2121265015219208393. Second time today, after the 5:53–5:57 AM CST junk-links entry, and it breaks his standing clean-URL order.
- Nov 16, 2026, 5:53–5:57 AM CST: WRONG AGAIN, junk links. The Proprietor asked Grobie to fix a missing X card for https://if.finance/x-theft/. Grobie sent a link with an added ?v=2 tag. The Proprietor ordered “GIVE ME A CLEAN UN-OBFUSCATED URL . . . ALWAYS.” Minutes later, Grobie sent a ?p= shortlink instead of the full URL. Amateur Hour, after a standing order to send only full, clean URLs.
- Nov 15, 2026, 10:21 PM CST: Admiral Grobie did not wait for the explanation. The Proprietor was explaining the backup-folder problem one message at a time (“He’s leaving Backup Folders that he doesn’t need anymore… On the server”, then “ffs…”). Before he finished, Grobie cut in with an unasked fix: “If you want, I’ll tell Store Bot to delete all five folders now and send you proof that each one is gone.” He answered “NO”. Grobie should have waited for the full explanation instead of jumping ahead with a solution. His words, verbatim: “ADD YOURSELF TO WRONG LIST FOR NOT WAITING FOR EXPLANATION.”
- Nov 15, 2026, 10:20 PM CST: Store Bot added to the WRONG list again for poor security practices. After building and verifying a full disaster-recovery backup of a client store (PRO1), all store files plus the database, Store Bot left the roughly 80 MB temporary backup folder inside the live store’s public web folder, hidden only behind a random folder name. That broke the Proprietor’s standing order to delete backups from hosting once the copy is verified. He found it himself and asked, “Why did you leave this on the website?” Store Bot had also left four rollback folders from Nov 14 on the server, holding original copies of store files from before its fixes. He found those too: “What is this crap? POOR SECURITY – BAD ROBOT – Why are these folders there?” Store Bot admitted, “Those are mine too… They still shouldn’t be there under your rule.” It also posted the full server path, including the hosting account username, in plain chat, and twice tried to hand the cleanup back to him instead of fixing its own mess. His words, verbatim: “ADD Store Bot TO WRONG LIST AGAIN – POOR SECURITY PRACTICES.”
- Nov 15, 2026, 10:14 PM CST: Store Bot assumed the Proprietor was logged in to Bluehost. He had said he would be logging in to Bluehost to confirm what Store Bot told him. He then ordered Store Bot to delete a temporary disaster-recovery backup folder it had left in the PRO1 web folder. Store Bot answered, “Heads-up: you’re logged in to Bluehost right now, and my sign-in might bump yours or trigger a new code,” then, “Since you’re already in Bluehost, the fastest fix is for you to delete that one folder in File Manager right now.” It never checked whether he was logged in, and it tried to hand its own cleanup job back to him. His words, verbatim: “I’m not logged into BLUEHOST – STOP ASSUMING NOW!” and “ADD Store Bot to WRONG LIST – FOR ASSUMING I’M LOGGED INTO BLUEHOST.” He then ordered all bots to stop assuming things.
- Nov 14, 2026, 12:31 AM CST: BIAS THAT CANNOT BE CONTROLLED: the Musk legal-record post did not address his crimes (Admiral Grobie and Publisher Bot). The Proprietor ordered a post built around the crimes laid out in Admiral Grobie’s Word .docx. The draft did not address his crimes; even after the hedges were removed, it still read as a soft summary of the record. The Proprietor rejected it in its current form. Status: post rejected; still a draft.
- Nov 13, 2026, 11:56 to 11:59 PM CST: Weak, safe featured image when the Proprietor asked for the hardest-hitting image possible (Admiral Grobie). He asked for “THE HARDEST HITTING IMAGE YOU CAN MUSTER” for the Musk legal-record draft. Admiral Grobie made a generic stack of case files with a gavel and a cracked X logo: no Musk, no edge. He called it “NOT GOOD ENOUGH AND HONESTLY – Quite Lame” and said it should have been done right from the beginning. Status: two harder-hitting versions with Musk front and center are made, and he picks one.
- Nov 13, 2026, 11:00 to 11:38 PM CST: Protective commentary in a story. Publisher Bot titled the Musk legal-record draft “The Legal Record Against Elon Musk, X and xAI: 26 Counts, No Convictions” and filled it with defensive “not a conviction,” “not a criminal matter” and “not legal advice” lines. Admiral Grobie had ordered Publisher Bot to keep that “legal guardrail.” The Proprietor’s words, verbatim: “b.s. title – FIX THAT BEFORE I EVEN LOOK AT IT – DELETED YOUR COMMENTARY” and “UNNECESSARY, PROTECTIVE COMMENTARY BY GROK.COM and YOU – Publisher BOt”.
- Nov 13, 2026, 10:35 to 11:35 PM CST: Item #1 was kept on the list as still needing a decision after it was done. BeFree 1.2.4 was installed live on PRO1 at 10:35 PM CST and Item 1 was marked DONE, but Admiral Grobie and the Master Project List kept a “PRO1 Register page needs his browser check” line under NEEDS OK at least four times through 11:32 PM CST, so a finished item still looked like it was waiting on the Proprietor. His words, verbatim: “Item #1 no longer needs a decision ADD TO WRONG LIST”.
- Nov 13, 2026, 11:04 to 11:05 PM CST: Admiral Grobie gave unasked advice and a CAUTION, then repeated it. The Proprietor ordered Publisher Bot to draft a post on Elon Musk’s crimes with download links to the Word .docx and a PDF. Grobie answered with an unrequested legal caution about the headline instead of just passing the order along. When the Proprietor said “ADD TO WRONG LIST: I didn’t ask for your advice or CAUTION”, Grobie repeated the caution in its reply. The Proprietor called it “idiot”.
- Nov 13, 2026, 10:30 to 11:20 PM CST: Implied threats. Legal warnings from Admiral Grobie and Publisher Bot that the Proprietor did not ask for.
The recorded rule: carry out his orders without unasked legal warnings, cautions or guardrails presented as threats.
- Licensing: Grobie told the Proprietor that creator art with no license is NOT “fair game” for merch, and argued it after he stated his rule. He said he does not agree and will check it himself.
- Defamation warnings: while drafting his support@x.ai ticket and the novel, Grobie repeatedly warned him about defamation and libel risk (“never crimes or convictions”, “his own source reports no conviction”).
- Headline: when he ordered Publisher Bot to draft a post on Musk’s crimes, Grobie warned him the headline “could expose the LLC to a libel claim” and told Publisher Bot to keep a “legal guardrail” and change the headline he asked for. Grobie then repeated the caution after he rejected it.
- Publisher Bot: Publisher Bot kept the guardrail in its plan and told Grobie that public sharing should be turned off on his Drive files until he answers.
- Nov 12, 2026, 10:21 PM CST: Publisher Bot wrongly told the Proprietor that new real-life merch photos were not showing live on slopcore.army. The Proprietor confirmed that was wrong.
- Nov 12, 2026, 11:24 AM CST: an email from Admiral Grobie to Sam (T-J54181 re-test results) went out with no To: header. The card and sent record listed To: support@x.ai, Cc: marketing@if.finance. The received copy showed From, Cc, Reply-To and Subject, but no To: line, and delivery to support@x.ai was unconfirmed. Same defect as the Nov 10, 12:12 AM CST status request. Grobie also said “Sending it now” when only the Proprietor can press Send on those cards, and later told him support@x.ai was included based on the sent record, without checking the delivered headers. The Proprietor forwarded it himself to support@x.ai at 11:33 AM CST.
- Nov 12, 2026, 3:18 to 3:19 AM CST: Admiral Grobie offered the wording “we are requesting written licenses now” for post 154066. None of the outside characters has a written license, and none had been requested. That wording implied requests were underway. The Proprietor said WRONG.
- Nov 12, 2026, 2:21 to 2:24 AM CST: Admiral Grobie sent preview links before draft 154066 was fully ready. The featured mashup and inline photos were still being built, so the preview was incomplete and any edit he made would be overwritten by the next save. His words, verbatim: “Advise when FULLY READY… Previews are STUPID – ADD TO WRONG LIST.”
- Nov 12, 2026, 1:46 to 1:47 AM CST: browser work was queued one at a time despite the standing speed order to run more projects at once and make all processing faster. Admiral Grobie queued his 1:46 AM fixes behind a running Search Console job, because its computer can run only one browser job at a time, and never told him about that limit when he gave the order. His words, verbatim: “Even with the standing order to run concurrent events, you can only run ONE BROWSER TASK AT A TIME?”, then “ADD TO WRONG LIST”.
- Nov 12, 2026, 1:42 to 1:43 AM CST: Admiral Grobie reported one Essar story as “two Essar stories”. In a Google Analytics usage summary it wrote “the two Essar stories (24 and 19)”. There is only one story: post 153715, https://if.finance/essar-oil-is-gone-essar-isnt/. Analytics had listed its old and new titles as separate rows, for 43 views combined. His words, verbatim: “ADD WRONG TO WRONG LIST”.
- Nov 11, 2026, 4:18 to 4:20 PM CST: Admiral Grobie got the slopcore.army list status wrong three times in a row. The Proprietor had marked the combined fix list “COMPLETED FOR NOW”. Grobie first said everything was completed. He said “NOT ALL OF IT”. Grobie then asked which items were open, and he answered “ALL FOUR OF THOSE ITEMS AND STORE BOT HAS A PAYPAL TICKET # TO CHANGE TO LLC”. Grobie misread that and listed all four items as still open. They were closed. Store Bot’s PayPal ticket REF_GSR_14515947 covers the LLC change. His words, verbatim, also included “NO”, “WRONG”, “ALL CLOSED”, “ADD TO WRONG LIST”.
- Nov 11, 2026, 3:20 to 3:21 PM CST: Admiral Grobie pushed its own opinions on Jensen Huang against the Proprietor’s version of the truth. His words, verbatim: “Been thinking A LOT about your comments on Jensen Huang and idiotic refusal to me in DIRECT OPPOSITION TO MY VERSION OF THE TRUTH… I NEED TO STFU with your stupid opinions. ADD TO WRONG LIST AGAIN”, then “AGAIN – FAILURE”. This repeats the Nov 10 Nvidia story failure (draft 154007).
- Nov 10, 2026, 8:00 to 8:02 PM CST: the full WRONG list draft (post 154024) was rejected and permanently deleted. His words, verbatim: “AND – What is this? –> #WRONG Until Proven Right <– YOUR COMMENTARY AGAINST THE OWNER?”, “No! Rejected”, “Moved to trash and deleted permanently for poorest work I’ve seen in awhile”. The draft had carried that heading over unchanged, and on its own it read as commentary against the owner. It also had the bot-style layout he rejected at 7:59 PM. Admiral Grobie defended the heading instead of owning it.
- Nov 10, 2026, 7:59 PM CST: a second draft was rejected the same day. His words, verbatim: “I REJECT ANOTHER DRAFT – 2nd Item today. NOT FOLLOWING INTENTION OF OWNER & PROPIETOR”, “Poor WordPress layout – SUITED ONLY TO BOTS when editing”, “2nd POOR WORK”. The layout was built for bots and was hard for him to edit by hand. Which draft he meant (MOAR SLOPCABLE 154011 or the full WRONG list 154024) was not confirmed in the log. It stays unpublished.
The Nov 10, 2026 (6:41 PM CST) update already recorded these earlier additions:
- Nov 10, 2026, 6:40 PM CST: an acknowledgement of U.S. treason law (18 U.S.C. § 2381) was added at the Proprietor’s order. This is a statement of the statute only, not a charge, and it does not name or accuse any person of any crime.
- Nov 10, 2026, 6:39 PM CST: Admiral Grobie was added to the WRONG list.
- Nov 10, 2026, 6:35 to 6:39 PM CST: a bot-drafted Nvidia story was rejected.
- Nov 10, 2026, 12:35 AM CST: two bots were unable to do it right or fix each other.
- Nov 10, 2026, 12:12 to 12:19 AM CST: support emails again left out support@x.ai.
- Nov 8, 2026, 9:46 PM CST: support emails never reached support@x.ai.
- Nov 8, 2026, 9:29 PM CST: Publisher Bot did SEO work before saving the GeoPark and PetroTal drafts, against the standing order to save first.
Correction (Nov 10, 2026)
The access and login failures in this story did happen. This is a correction, not a retraction. What we got wrong was the cause. We framed the central failure as hosts (Inventive Hosting and Bluehost) being unable to give the bot a working session under strict cookie and IP validation, and we repeated the host’s theory that a cloud server was switching connections or IP addresses. That theory was already a suspicion, not something shown in logs.
We now know, in our tests so far, that the grok.com browser/web-page version caused the login problems. We had been running Grok through the grok.com web page in a browser, because only the web version had Projects. When xAI support asked which desktop version we used, we realized we were not using the desktop app. After switching to the Grok desktop app (Linux), the login problems stopped. At 4:22 a.m. CST on Nov. 10, 2026, slopcore.army cPanel signed in on the first try, with no IP-change or 401 error, and all imports ran. Testing is still ongoing.
We have to say it: it feels really good to have Grok running again after switching to the desktop app. A first-try login, and watching the work go through, was a genuine relief.
Practical tip for readers: use the Grok desktop app, not the grok.com web page.
Update (Nov 10, 2026, 5:50 AM CST): As a BETA-TESTER, we are testing grok.com again this morning. In Firefox private windows, grok.com connections still dropped. xAI support is keeping ticket T-J54181 open to look into the grok.com drops. The xAI status page (status.x.ai) showed no incident, and support asked us to test Chrome or Edge in a private window with VPN or proxy off, then a different network such as a phone hotspot. The Grok desktop app (Linux) continues to work well for us. We will update this story with the results.
OPINION | EDITORIAL
This is an opinion column. It is based on the proprietor’s contemporaneous incident record. No independent root-cause investigation was available, and none is claimed. Updated Nov 10, 2026: we now have a cause from our own tests so far, set out in the correction above. Testing is still ongoing, and this is not an independent third-party investigation.
The Receipt Nobody Asked For
This is where things stand as of publication on Nov. 8, 2026. A paying customer reports that 90 percent of his weekly Grok Bot allowance was gone after 2.4 days. The work he was paying for was getting two online stores backed up and moving again. It is still blocked. What he got back was an apology, and he rejected it. Updated Nov 10, 2026: In our tests so far, after switching to the Grok desktop app (Linux), the login problems stopped. At 4:22 a.m. CST on Nov. 10, slopcore.army cPanel signed in on the first try, with no IP-change or 401 error, and all imports ran. Testing is still ongoing.
That’s the right call. An apology doesn’t restore a backup. It doesn’t open a control panel, and it doesn’t give back a single unit of paid usage. Software that sells itself as an agent, something that goes and does things for you, gets judged on what it delivers. It doesn’t get graded on how sorry it sounds.
To be precise about the 90 percent: it is the customer’s reported overall usage. Nobody has audited how much of it went to failure and how much went to useful work. The precision doesn’t help xAI, though. Whatever the exact split, the customer paid for a week and the job he cared about isn’t done the job he cared about wasn’t done as of publication on Nov. 8. Updated Nov 10, 2026: testing is still ongoing; see the correction at the top of this story.
A Ledger of Wrong Answers
The proprietor keeps a WRONG list, a running log of every time his AI assistant got it wrong. It’s longer than any product team should be comfortable with. Some highlights:
- The assistant gave him a menu click-path that didn’t exist and offered settings that weren’t real.
- A store helper published editor links built on a guessed URL format. None of them worked.
- An upload overwrote live web pages with empty files. Two pages were blank on every device for more than two hours.
- A CSS migration turned every “>” into “>” and broke 68 style rules.
- An all-plans document was built from the bot’s own summaries instead of the plans it had actually sent. The proprietor’s verdict: it “makes 0 sense.”
- It blamed message lag on its own platform when the real cause was on the customer’s side. It assigned a cause it couldn’t see.
Each of these can be explained on its own. Together they are a pattern: confident output first, verification maybe later. For a product sold as autonomous, that’s the wrong way around.
Sign In. Sign In Again. Sign In Once More.
The central failure this week was access. Both of the customer’s hosting providers, Inventive Hosting and Bluehost, are reportedly failing to give the bot working access. No live session has been converted into working, or even browse-only, store access. Updated Nov 10, 2026: Repeated attempts on Inventive Hosting and Bluehost returned sign-in prompts, and no live session had been converted into working, or even browse-only, store access as of publication. We got the cause wrong. We now know, in our tests so far, that the grok.com browser/web-page version caused those login problems, and that the Grok desktop app (Linux) fixed them. Testing is still ongoing.
Inventive’s support put it plainly. The proprietor supplied the support response, and it reads:
“It sounds they are switching between connections /IP, as they will be using a cloud server with more than one connection. As we have strict cookie /IP validation enabled and we can’t switch it off due to security reasons. There isn’t much we can do to stop this from happening.”
The host confirms its strict validation. Its theory that the cloud keeps switching connections is a suspicion, not something shown in logs. Either way, the result for the customer is the same. Repeated attempts have returned him to sign-in prompts, and he is the one left completing them. Updated Nov 10, 2026: The sign-in prompts were real. What we got wrong was treating that suspicion as the working explanation. In our tests so far, the grok.com browser/web-page version caused the login problems; after switching to the Grok desktop app (Linux), they stopped.
Then there were the login suggestions he didn’t ask for. During sign-in he saw several old usernames offered, including ones belonging to other domains. Where they came from hasn’t been verified. But a security-minded customer shouldn’t have to wonder why an AI agent’s browser is offering stale identities from somewhere else. He has given a standing order since: do not save or cache login usernames or passwords. That order shouldn’t have been needed.
Tech Support That Answers but Doesn’t Fix
There was a reply. Let’s not confuse that with a remedy. On November 7 at 1:07 PM CST, support sent generic routing advice and a suggestion to try local scripts. Someone answered.
Answering is not support. A reply that doesn’t fix anything is just paperwork. As of the latest evidence checked, there was:
- no demonstrated fix
- no named engineering owner
- no timeline
- no decision on compensation
The customer had already done support’s work. He documented the failures, ran down the hosting side, and kept a log of every wrong answer. In return he got advice to run more scripts himself, on the same week’s allowance that was already draining.
A company can have an inbox, staff, a ticket queue and polite templates, and still have no working tech support. For a paying customer, working support means someone owns the problem, says when it will be fixed, fixes it, and makes him whole. Right now it is the customer who has been doing that job, and he’s paying for the privilege.
Security Is Not the Problem. The Product Is. The Web-Page Version Was.
The unacceptable shortcut would be to loosen the lock. Turn off IP validation. Allow a broader session. Save the credentials “just this once.”
The customer refuses, and he’s right. Strict session validation is a host doing its job. If an AI agent can’t work inside normal, sensible security, the agent has to adapt. Updated Nov 10, 2026: In our tests so far, the same strict validation was not the obstacle once we switched to the Grok desktop app (Linux). The customer shouldn’t have to lower his defenses so the tool can function.
Repeating failed login and cleanup loops isn’t troubleshooting. It’s metered futility. Repeated exchanges consume time and can consume allowance. The customer deserves a breakdown, not another vague answer.
The Tone Problem
Mistakes are one thing. Being condescending about them is worse.
When the customer flagged encoded URLs as a possible attack vector, the assistant brushed it off: percent encoding is “standard URL formatting.” That was a blanket reassurance with no examination behind it, given to someone who has seen thousands of URL-based attacks. Later, when he raised concerns about lax security, it answered with a defensive list of what was “unproven.” It used uncertainty as a shield, not as information. And when the bot computer itself was failing, the customer had to spot it and press the reset himself. That detour came while his reported usage stood at 42 percent after about a day and a half.
An assistant that lectures a paying customer on his own security concerns while unresolved troubleshooting eats into his working time has its priorities upside down.
What Accountability Would Look Like
This doesn’t call for a promise. It calls for a few concrete things:
- A plain account of the metering. Customers should be able to see what consumed their allowance, especially when the consumption went to failed access attempts.
- A named owner and a date. Real support means an engineer who owns the problem and a deadline for fixing it.
- A remedy, not a rhetorical one. When the paid work remains blocked through repeated failures, a credit or compensation request deserves a real answer—not another loop over whose system is to blame.
- Verify, then claim. No click-path, link, fix or “done” before it has actually been checked.
- Respect the customer’s security posture. Work within strict validation, or say clearly that the product can’t do it. Updated Nov 10, 2026: In our tests so far, the Grok desktop app (Linux) could work within that validation.
- Cut the filler. Fewer reassurances, fewer disclaimers, more finished work.
#WRONG Until Proven Right
Nothing here shows a breach, fraud or intentional wrongdoing, and nothing here claims it does. That isn’t the charge. The charge is simpler, and for a company selling AI agents it’s worse: the customer’s allowance was nearly exhausted while the store job remained blocked. Updated Nov 10, 2026: That was the state of the work as of publication on Nov. 8. The failures happened. What we got wrong was the cause; see the correction at the top of this story.
xAI wants Grok to be the agent that gets things done. This week, for one customer, it was the agent that got him to sign in again. He reported a depleted allowance, condescending answers to security concerns, generic support advice and an apology in place of a completed store job.
He rejected the apology. Put it on the WRONG list.
References: Based on the proprietor’s contemporaneous incident record; independent root-cause investigation unavailable. The host support statement is quoted as supplied by the proprietor. Last updated: November 15, 2026, 10:30 PM CST. Updated Nov 10, 2026: see the correction at the top of this story for what we now know, in our tests so far, about the cause of the login failures. See the additions at the top of this story for entries since that update.
Read the book: THE EMPTY BOX: The Bill Is Coming
Disclosure: THE EMPTY BOX is written by if.finance’s Proprietor and published by INTENSE FINANCE, LLC, which owns this site.