Elon Musk said on X on October 15, 2026 that SpaceX will "make you whole" if Grok Bot makes a mistake with a purchase or leaks a user's card. The statement was a reply to a user who said Grok Bot had booked a flight after being shown a photo of a credit card, and it followed an earlier post in which Musk invited people to buy things by dropping a picture of a card into the chat. It was an informal post, and the terms of service that SpaceXSI last updated on October 12 say something different. This article lays out both texts side by side, what is and is not known, and why the gap matters for anyone letting an AI agent pay for things.
xAIWhat was said, and when
According to The Next Web's account, Musk's sequence of posts ran as follows. First he wrote that to buy things users could "just take a picture of your credit card, drop it in the chat," and said Grok Bot would find the best deal and order it. Then, after a user reported that Grok Bot booked a flight from a photo of their card, Musk replied: "If Grok Bot makes a mistake with a purchase or leaks your card (never happened yet), SpaceX will make you whole." Later he posted that he had asked Grok Bot to research Lego Star Wars kits for young children and order them, calling it "as easy as that." The post passed 27 million views, and the order was $241.34 for five sets on Lego's website, which Musk later said arrived.
Two details around the demo are worth recording as reported. A reply pointed out that Musk had exposed his Grok Bot email, delivery address and order number in the screenshot; Musk said the address was already widely published and is where his security team receives mail. And a user claimed Amazon was blocking the agent, to which Musk replied that Grok "will find another way to buy stuff." The Next Web notes it did not verify the Amazon claim. A separate analysis on this site covers how Amazon has handled third-party agents.
Grok Bot launched in August as part of the AI company since renamed SpaceXSI. It is not on the free plan. The X Subscription launched on October 12, and its $30 per month Premium tier is described by SpaceXSI as including SuperGrok, Cursor Pro and X Premium+, products that The Next Web says cost $90 per month combined before.
Two texts that do not say the same thing
The terms of service, last updated October 12, say Grok can take "Agentic Actions" on a user's behalf, including with "financial institutions." Their key disclaimer reads: "We are not responsible for User Content or Agentic Actions." The Next Web notes that Musk's pledge is not mentioned in the terms.
A community note on one of the posts, as described in the same report, makes further claims about the terms: that xAI warns users not to put personal information in prompts, that card photos may contain such information, that the terms let SpaceXSI store user data and credentials in its cloud, and that financial transactions need human review under those terms. The article presents these as the note's claims, not as independently checked, and so does this one.
| Source | What it says | Binding? |
|---|---|---|
| Musk's reply on X (Oct 15) | SpaceX will "make you whole" after a purchase mistake or a card leak | Not stated; no scope, claim process or conditions published |
| Terms of service (updated Oct 12) | "We are not responsible for User Content or Agentic Actions" | The contract users accept |
| Community note (as reported) | Terms require human review for financial transactions; warns against personal data in prompts | A claim about the terms, not verified |
A public pledge and a contractual disclaimer can both be true at once, but only one of them tells a user what happens when something goes wrong.
Analysis
The questions the pledge leaves open
The pledge covers two things, a "mistake with a purchase" and a card leak, and defines neither. Reading the plain words, several situations fall in a grey zone, and none of them is answered in the reporting:
- Wrong item or wrong quantity. Is an order that matches the user's request but disappoints them a mistake?
- Unintended merchant or price. If an agent finds a "best deal" on an unfamiliar seller's site, does the pledge cover the difference?
- Downstream costs. A flight booked in error may carry change or cancellation fees, or trigger other costs.
- Third-party fraud. If a card detail is taken from a merchant the agent used rather than from SpaceXSI's systems, whose "leak" is it?
- Process. There is no published claim route, time limit or evidence standard.
None of this means the pledge will not be honored. It means a reader cannot tell from public text how it would be applied, and the report notes SpaceX did not respond to Business Insider's request for comment.
Why the design matters more than the quote
Handing a card photo to a chat agent is one of several ways agents can pay. A neutral way to compare them is by how much authority the agent holds and where the human checks sit.
The point of the scale is that the risk a user carries depends on design choices that are easy to miss: whether the agent sees the full card number or a tokenized, limited credential, whether each purchase needs an approval, and whether the merchant, amount and delivery address are shown before the order is placed. A reimbursement promise is a remedy after the fact. Spending limits and confirmations are controls before it. Other financial firms have taken the opposite stance on agent access; a related article covers why JPMorgan and Schwab block outside agents from moving customer money.
How to read this story
- What is established: Musk made the pledge in a post; the terms disclaim responsibility for Agentic Actions; Grok Bot placed a $241.34 Lego order.
- What is not: the scope of the pledge, whether it will be written into the terms, how financial transactions are reviewed, and whether Amazon is in fact blocking the agent.
- What to watch: any amended terms or published claim process, and whether card networks, issuers or regulators comment on agent-initiated purchases. Those are the places where liability gets settled in practice.
For individuals and teams trying agents with real money, the safe pattern is the boring one: use limited-purpose or virtual card numbers, set per-transaction caps, and require confirmation above a threshold. Tools that keep the agent's model choice separate from the payment controls, which is the approach platforms such as Metir take with approvals on actions, make it easier to apply the same rules whichever model is doing the shopping.
Sources:
- Musk says SpaceX will cover Grok Bot purchase mistakes and leaked cards (The Next Web)
- Elon Musk says SpaceX will pay back anyone whose Grok AI agent makes a rogue purchase (DNYUZ, Oct 15, 2026)
- Elon Musk, SpaceX will pay users back for Grok purchase mistakes (Business Insider, Oct 2026)
- Everything That Happened in AI Today, Monday, October 19 (The Neuron)
Image credits
Header and in-body image: Elon Musk, March 28, 2024, by Wcamp9, Wikimedia Commons, CC BY 4.0 (source); illustrative.