Elon Musk is once again drawing attention with bold statements. This time, the focus is on the security of the AI agent Grok, which has been entrusted with managing users' finances. The head of xAI publicly promised to compensate any losses if the bot makes a mistake when handling a client's funds. It sounds like a revolution in trust in artificial intelligence, but upon closer inspection, things are not so clear-cut.
An experiment with a bank account and Musk's promise
The reason for the discussion was a post by a Tesla and xAI investor under the handle Teslaconomics. He asked a provocative question: has anyone already connected Grok to their bank account? After all, such an agent could potentially track expenses, pay bills, and identify suspicious transactions. One user expressed concern that an AI with access to money could cause trouble. However, Musk did not flinch, declaring full confidence in the product and a willingness to cover losses in the event of an error.
Legal nuances: A $100 guarantee
The beta version of the product saw the light of day on August 11. xAI emphasizes that each agent operates autonomously on its own cloud server, working even while the owner sleeps. However, contrary to the bold statements, xAI's user agreement spells out the exact opposite logic. The company's maximum liability is limited to $100 or the amount of fees paid, whichever is greater. This looks more like a mockery, given that access to the bot under the SuperGrok plan costs $30 per month, and an annual subscription is $360. Funds lost from an account could be orders of magnitude higher.
The key point: as long as Musk's guarantee is not legally formalized, any compensation remains solely at his discretion. This promise is nothing more than a gentleman's agreement, not backed by formal obligations.
Real risks and attacks on AI agents
The situation is further aggravated by regulatory risks. In the United States, Regulation E protects consumers from unauthorized debits. However, if the owner themselves granted the bot access to the account, such transactions cease to be considered unauthorized, and standard protection mechanisms will not apply. Skeptics also point to recent incidents demonstrating AI's vulnerability. In May, a malicious NFT "hid" instructions that forced the agent to transfer funds — a classic example of a "prompt injection." As a result of the attack on the Bankr wallet linked to Grok, about $150,000 was withdrawn. Later, roughly 80% of the stolen funds were recovered, and Bankr promised to fully reimburse losses for 14 other affected wallets. Notably, not a single proven case of Grok making an error with a real bank account has been recorded so far.
Analytical commentary: This experiment is not just a test of technology, but a powerful marketing move to integrate AI into everyday financial management. The very first real failure will be the litmus test. The choice for Musk is obvious: either silently return all funds to save face, or publicly compensate only that very $100, exposing the gap between bold words and legal reality. I would not advise entrusting critical financial operations to an AI agent until its responsibility is backed by real guarantees, not just tweets.