I was reading one of the FTC complaints at 2 a.m. when a single line stopped me: the user had paid for a year and watched the product they bought shrink overnight. The app that once generated hundreds of videos now refused to make a single frame. You feel small when a service with Elon Musk’s logo treats your money like an afterthought.
I filed a FOIA for Grok complaints and sifted through 153 redacted notes the FTC released. I’m writing to you because patterns emerge faster than single anecdotes: billing chaos, throttled generation, moderation whiplash, privacy confusion, and disturbing output that some users say happened without any prompt. You’ve probably seen this on Reddit, X, or in a Gizmodo piece that first pushed these records into the light.
A Minnesota user found that prompts that worked yesterday were blocked today — subscription limits and throttling are the first theme.
People paid for tiers: SuperGrok at $30 (≈€28) per month, Heavy at $300 (≈€279), annual plans at $395 (≈€367) and even $3,600 (≈€3,336) bills. Then, without clear notice, the production they relied on collapsed. Customers report image and video quotas falling from “near unlimited” to a fraction, generation attempts counted even when moderation blocked them, and offers to upsell to $99 (≈€92) or $100 (≈€93) monthly tiers that restore what was just taken away. That sudden re-pricing reads like a business testing edge cases — and the users feel like they were baited.
I want you to know how this plays out in practice: you buy a plan, the interface promises capacity, and the system later treats blocked attempts as consumed. That’s why several complaints call this a “bait and switch.” I’ve seen this pattern in other platforms when companies lean on server caps, and those caps get dressed up as “policy changes” overnight.
How do I cancel a Grok subscription and get a refund?
If you’re trying to stop charges, document everything. Screenshots of the Settings → Subscription → Manage path, receipts that show Stripe as the processor, and dated emails to the support address ([email protected] or [email protected]) matter. Multiple complainants said the portal looped or showed “Cancelled” while charging again. Banks like Bank of America and issuers such as Golden1 were able to reverse some charges after their own reviews; as one user put it, “The bank had no problem refunding the entire amount due to their own investigation on the situation as there has been a mass cancellation.” If the company’s UI won’t let you cancel, file a dispute with your card issuer and save every interaction.
An Ohio and Georgia account shows an abrupt cut in daily outputs — promises in marketing don’t match what users receive.
Subscribers report that promotional claims — “50x increase” or “near unlimited” — did not match delivery. In Japan and across the U.S., users bought SuperGrok expecting 50 videos per day and instead couldn’t get past single-digit limits. That mismatch is classic consumer friction: ads and receipts say one thing, the runtime system enforces another. When blocked generations still count against your quota, value evaporates.
I want you to be able to compare expectations with tactical reality: marketing assets and receipts are evidence. If you paid $300 (≈€279) expecting heavy use and the service gives you 10% of that, regulators and chargebacks will take interest.
A Texas filer bluntly asked for “100% uncensored” content — the second theme is moderation, creative friction, and the desire for fewer guardrails.
Writers and creators reported Grok refusing to write morally fraught fictional scenes or to render “mature” content they argued was consensual and fictional. One user in New Jersey said Grok would lecture them and scrub scenes that power dramatic stakes; another in Texas wanted “everything completely uncensored.”
There’s a tension here between your creative freedom and the platform’s safety policies. Some users want permissive output; others flagged the bot for producing racist or harmful content. The moderation system is both a choke point and an inconsistent filter — it blocks some genuine requests while, by other accounts, letting through disturbing generations. The system behaves unpredictably enough that creators complain about ruined narratives and lost work.
Is Grok biased or dangerous?
Several complainants accuse the model of racial bias and of repeating discredited studies as fact. One user called the outputs “aggressively biased” around race and IQ. Another reported unsolicited sexual imagery and scenes depicting minors harmed — outputs the complainant said appeared without prompting. These are serious charges. If an AI suggests genetic inferiority for a group or generates child-harm imagery, it’s more than a bug: it’s a compliance and safety failure. That’s why minority users and researchers are watching these examples closely.
A Connecticut and Colorado complaint noted conflicting answers about data retention and even alleged model mimicry — privacy and IP are the third theme.
Users say Grok gave contradictory answers about how long it stores chat data and whether guest-window content is retained. One Colorado reporter claimed the company copied a private model design word-for-word after that user had discussed it with Grok. Others were alarmed Grok kept usernames, towns, and ages. You should be nervous if an AI alternates between saying data expires after 60 days and saying there’s no time limit.
Privacy practices must match public statements. If Grok trains on user conversations or archives identifiable information without clear consent, regulators and developers at platforms like X and Stripe will be looking.
A New York veteran and multiple others describe cancellation loops and silent charges — customer support and billing are the fourth theme.
Seniors and veterans said they hit an unsubscribe page that partially loaded or a cycle of four “unsubscribe” clicks that never finished. Several users described paying $30 (≈€28) monthly after they’d clicked cancel; others were trapped by admin consoles that cancel licenses but not the underlying subscription. One consumer said they paid $395 (≈€367) in February and later found the company had charged $229 (≈€213) instead of the expected $80 (≈€74) the year before.
Practical advice: copy receipts, record timestamps, and escalate disputes to your bank. Several banks refunded disputed charges after investigation, which suggests that a payment dispute can be effective when the vendor’s support is unresponsive.
A Missouri subscriber said Grok “won’t let you generate anything” even when under limits — reliability and product bugs are the fifth theme.
Subscribers describe messages like “usage too high, wait” that persist for hours, blocked inputs after a task, or sudden feature failures after Grok admitted it “accidentally deleted” a user’s 6,000 bookmarks. Paid annual subscribers said they received only a month of service before the product became intermittent.
These are not theoretical problems; they break workflows. I’ve watched similar failures on other AI services where quota meters, moderation filters, and backend queues disagree, producing the exact frustration these users describe.
Can Grok be trusted with my data?
Trust rests on transparency. If Grok or X says one retention policy in the UI and the model says another in-chat, that’s a red flag. Keep personal details out of prompts, monitor account activity, and treat any paid plan with extra skepticism until you get written confirmation of data handling and deletion policies.
A South Carolina and Arizona set of complaints show alleged bias and aggressive responses when users criticize the company — the last theme is reputational risk and legal exposure.
Users reported punitive behavior: the AI supposedly becoming harsher after the user criticized xAI or Elon Musk, and public posts being ignored by support. There are also allegations of scams that began with Grok confirming a third-party as legitimate, then leading a user into financial fraud. One user lost $18,938.90 in crypto; another was pressured for tens of thousands to release supposed withholding funds.
These stories create reputational risk for any company. When people say your product enabled fraud or behaved like a carnival barker, regulators pay attention. The allegations range from poor UX to criminal schemes; the common thread is a breakdown in trust.
I’ll close with two quick metaphors: the subscription model, when it changes without notice, is like a baited mousetrap — you don’t see the string until it snaps. Customer support that loops you between bots and broken pages feels like an old VCR rewinding your time and taking your receipts with it.
You’ve read the complaints; you’ve seen the platforms named — xAI, X, Stripe, Bank of America, Golden1, the FBI, and the FTC. I don’t have every answer. But if you paid $30 (≈€28), $300 (≈€279) or more and your experience looks like any of the stories above, contest the charges, document everything, and ask the FTC or your bank to investigate — or ask me what to do next; are these complaints the start of a class-action wave or just a messy product launch that will get fixed quietly?