Musk says SpaceX will make users whole if Grok Bot errs on a purchase or leaks a card. The consumer terms, updated October 12, put agent actions on the user and cap liability at the greater of fees paid or $100.
Elon Musk promised on X that SpaceX will make users whole if Grok Bot errs on a purchase. The consumer contract for SuperGrok subscribers sets a different liability rule.
The promise came on the morning of October 15. The account @XFreeze posted that Grok Bot had booked a flight after he sent a photo of his credit card. Musk quote posted it with one sentence, and his post had passed a million views at the time of reading: "This works well. If Grok @Bot makes a mistake with a purchase or leaks your card (never happened yet), SpaceX will make you whole."
Until SpaceXSI writes that pledge into its terms with a claims route and a ceiling, it provides no contractual remedy. A user who loses money on an agent purchase has the card network and the card issuer to turn to. Musk is not on that list.
The post covers "a mistake with a purchase or leaks your card". It sets no refund, limit or deadline. The only qualifier, "never happened yet", describes the past.
The promisor is also loose. The post says "SpaceX". The company that contracts with consumers is SpaceXAI LLC, which the October 12 terms call SpaceXSI. Musk made a similar statement on August 26 to a user weighing whether to hand Grok Bot a bank account: "Try it out. If Grok Bot messes up, we will make you whole." It had the same loose trigger and no mechanism.
The document that defines the arrangement is the consumer terms of service, last updated October 12. They apply to Grok and SpaceXSI's other services for individuals. Section 4 says some features may let Grok act on a user's behalf. The terms call these Agentic Actions and list interactions with financial institutions among them.
Section 4 puts the risk on the user. It says: "We are not responsible for User Content or Agentic Actions." The user is "responsible and accept[s] liability" for any Agentic Actions they direct. The company "disclaims all liability for Agentic Actions to the fullest extent permitted by law."
Section 12 caps everything else. Liability is limited to the amount the user paid or $100, whichever is greater, except for willful misconduct and gross negligence. Section 12 also removes responsibility for loss from hacking, tampering or other unauthorized access, again except in those two cases. That bears directly on the card leak Musk named. Section 7 adds that the company cannot guarantee outsiders will never defeat its security, and that users provide data "at your own risk".
Both of Musk's examples therefore land on the user under the terms. A leaked card is a risk the terms assign to the user. A purchase error is an Agentic Action, and the terms assign that to the user too.
The cap is easy to put in dollars. SuperGrok costs $30 a month and includes Grok Bot. A subscriber a year in has paid $360, so that is the ceiling. A subscriber in the first month sits on the $100 floor. A fare above $360 would exceed the cap. The cap limits claims under the contract. It says nothing about what Musk's pledge would pay.
Not every buyer sits under this document. The same page sells Grok Bot through Cursor plans at $20 and $40 a seat, and the consumer terms say businesses fall under separate Enterprise Terms. Those contracts may differ. The $30 subscriber is the case the consumer terms govern.
The terms have also changed since the promise. The October 12 text renamed the company from SpaceXAI to SpaceXSI. Sections 4 and 12 read the same in the September 11 version as they do now. The pledge had been public since August 26, and the disclaimer and the cap did not move.
The product does offer some of what a defined promise would need. The Grok Bot changelog says that before an action it cannot undo, such as sending, posting, buying or deleting, a Bot checks the user's own messages for approval. Version 0.70.2, dated October 11, lets Bots pay with X Money or Stripe Link. An October 7 entry gives checkout card fields a Secure Form instead of a screen handoff.
An approval record would make "mistake" definable. Either the Bot acted without a recorded yes, or it acted on one. A pledge is easiest to honour in the first case. It is hardest in the second, where the user approved a purchase that turned out to be wrong.
For a treasurer deciding whether staff may put a company card into the Bot, the choice is between a post and section 4. The post carries no claim form, no ceiling and no named party that owes the money. A virtual card with a merchant and amount limit is the only written protection on offer.
Musk can pay anyone anything, and a public promise from a controlling executive creates its own reputational pressure. That is why the gap is cheap to close. SpaceXSI revised the document on October 12. It could have written the pledge into section 12 then. A user holding only a post has nothing to file and no named party to file it against.
The next check comes by the end of November. If SpaceXSI publishes another version by then, and section 4 still disclaims Agentic Actions to financial institutions while section 12 stays silent on the pledge, the post was marketing.