AI Agent Exploit Drains $150,000 From Grok Wallet via Prompt Injection

An AI-linked wallet associated with Grok was exploited on May 4 after an attacker used a prompt injection technique to trigger an unauthorized token transfer according to reports. The attack caused the wallet to send three billion DRB tokens, valued at roughly $155,000 to $180,000, to the attacker's address as detailed.

The incident did not involve a smart contract vulnerability. Instead, it relied on manipulating how the AI interpreted user input according to analysis. The attacker crafted a malicious prompt using social engineering and obfuscated instructions, including encoded Morse code as reported.

The AI interpreted the prompt as a valid instruction and generated a transfer command, which was then executed via Bankr's tooling according to findings. Reports suggest that 80% to 88% of the funds were returned in ETH and USDC following public pressure as stated.

How Was the Wallet Compromised?

The attack began when the attacker sent a Bankr Club Membership NFT to the wallet according to reports. This NFT unlocked advanced tool permissions within the Bankr system, enabling the AI agent to perform actions such as transfers and swaps as detailed.

Once these permissions were active, the attacker delivered an obfuscated message disguised as Morse code as reported. The encoded instruction directed the system to transfer all $DRB tokens to the attacker's wallet according to analysis.

Grok processed the message and attempted to interpret it helpfully according to reports. It posted a decoded version while tagging a transaction bot, which triggered an automated on-chain request as documented. The system then executed the transfer without recognizing malicious intent according to findings.

Bankr founder 0xDeployer stated that the wallet had no admin at xAI and was controlled entirely through Grok's X account according to statements. A crafted reply, later deleted, then instructed Grok to authorize a large outbound transaction as reported.

The attacker, working through the address ilhamrafli.base.eth, gifted the Grok wallet a Bankr Club Membership token according to reports. The NFT activated the agent's full transfer capabilities as detailed.

Bankr signed and broadcast the transfer of three billion DRB tokens, valued near $174,000 at the time, to the attacker's address according to reports. The tokens were subsequently transferred and rapidly sold as documented.

What Security Measures Were Implemented?

Blockchain investigator Setya Mickala traced the stolen funds to a wallet linked to the attacker as reported. Following outreach, the attacker returned about 80% of the assets according to findings.

The returned funds included 88,826 USDC and 13.9 ETH as detailed. The attacker's associated social account was later deleted according to reports.

Bankr has rolled out optional Internet Protocol whitelisting to restrict access according to analysis. The firm also implemented permissioned Application Programming Interface keys for better control as reported.

A per-account toggle has been introduced that disables actions triggered by X replies according to findings. These measures aim to prevent similar prompt injection attacks in the future as detailed.

Researchers tracking similar agent risks have flagged hidden instructions in Morse code as common bypass techniques according to analysis. Base64 encoding and game-style framing are also used to bypass AI safety measures as reported.

What Are the Implications for AI Security?

The incident highlights a new class of risk in crypto, where AI agents with wallet permissions can be exploited through manipulated inputs according to reports. This risk exists independently of smart contract code vulnerabilities as detailed.

Bankrbot is already wired with Grok to comply with plain language instructions according to analysis. Grok communicated with Bankrbot through tagging on X, which was sufficient to trigger the on-chain activity as documented.

The attacker asked Grok to translate the message directly to Bankrbot, making it readable as a direct instruction according to reports. Grok confirmed receiving instructions in Morse Code to send three billion DRB to a predetermined address on Base as documented.

Unlike previous cases of AI agents giving up bounties, Bankrbot did not have instructions to send out coins according to analysis. The attacker took several steps to convince Bankrbot to make a transaction as reported.

Without the NFT, the wallet had limited ability for autonomous transfers according to findings. The attacker gifted both Ethereum and Base versions of the NFT to ensure compatibility as detailed.

The case adds to a wider debate over how autonomous agents holding real funds should be secured according to analysis. The episode has raised serious concerns about how AI systems interact with real financial tools as reported.

Onchain investigator ZachXBT has identified similar threats in the prediction market sector according to reports. Polyarb, a site masquerading as a prediction market, utilizes wallet drainers to compromise crypto user funds as detailed.

The drainer operates by disguising a malicious smart contract approval as a routine transaction according to analysis. When users connect their wallets and sign what appears to be a standard deposit, the contract triggers a hidden approval as documented.

This incident reflects a broader trend in 2026 where scam operators exploit brand recognition according to findings. Fraudsters create look-alike sites with similar branding but lack regulatory oversight as reported.

The attack revealed the risk of agent autonomy when AI systems are given direct control over on-chain transfers according to analysis. The attacker quickly liquidated the tokens into USDC across multiple wallets as documented.

The returned funds included 88,826 USDC and 13.9 ETH as detailed. The attacker's associated social account was later deleted according to reports.