Blockchain investigator Specter flagged the theft on June 20: JaredFromSubway.eth, a second-iteration MEV bot operated under a deliberately offensive pseudonym, lost approximately 1,475 WETH, $2.9 million USDC, and $2 million USDT. The operator had built wealth across two bots and 6.4 million transactions by front-running and back-running user trades in the mempool—a practice known as sandwiching.
The exploit worked by design flaw, not brute force. Yearn developer Banteg reconstructed the attack: over 97 consecutive blocks, the attacker dangled small but real profits from fake token pairs. Each time JaredFromSubway's bot took the bait, it signed contract approvals to spend its WETH, USDC, and USDT. The bot never consumed those tokens during the profitable sandwiches, and crucially, never revoked the approvals. The attacker then swept everything in a final harvest transaction, using the pre-approved spend limits to drain real funds.
This was not a momentary lapse. The campaign required patience and precision. The attacker had to create fake-DEX arbitrage opportunities plausible enough to trigger the bot's logic, space them across blocks to avoid detection, and wait until the approvals accumulated into a haul worth claiming. That sequencing matters: it shows the attacker understood both the bot's behavior and Ethereum's contract mechanics well enough to exploit a common but avoidable vulnerability.
JaredFromSubway's operator responded on-chain with a ransom offer. In a transaction data message to the attacker, signed by the bot's wallet, the operator conceded "Well played" and offered 2,150 ETH—roughly 50 percent of the stolen funds—for their return within 48 hours. The message appended a legal threat: "Otherwise we will pursue all available legal and law-enforcement remedies."
and offered 2,150 [ETH](/coin/eth)—roughly 50 percent of the stolen funds—for their return within 48 hours. The message appended a legal threat:
The loss triggered a scavenger pile-on. A scammer using the X handle "jaredsmev" fabricated bounties of $1 million to $7.5 million, inflating the total loss to $15 million in false claims. Cointelegraph amplified one of the scammer's posts to 2.9 million followers before deleting it. Meanwhile, other users claimed to be victims of JaredFromSubway's own sandwich attacks and reached out on-chain requesting reimbursement, with one calling the attacker a "Robin Hood in a White Hat." The irony was not subtle: the bot that had profited by extracting value from ordinary traders was now being extracted from itself.
The attack exposes a gap in common bot security practice. Token approvals are a necessary evil in DeFi—contracts need permission to move user funds—but many bots set unlimited or high approval amounts and forget to reset them. JaredFromSubway's operator treated approvals as temporary, revoking none after each transaction. Against a patient, methodical adversary, that assumption cost $7.5 million.