Bitquery counted 6.85 million successful sandwich attacks on Solana over 11 months, with 96% targeting Pump.fun's launch curve and PumpSwap. The $29.7 million estimate is gross proceeds, not net profit; wider slippage settings were associated with greater exposure.
A swap can go through and still leave a trader with fewer tokens than expected. Bitquery estimates sandwich bots generated $29.7 million in gross proceeds on Solana over 11 months. That is not net profit, and blockchain records alone cannot establish who controlled every wallet or what each operator knew.
Where attacks clustered
In its Solana investigation, Bitquery identified 6,850,144 successful attacks by 2,545 bot wallets. The attacks affected 2,154,994 distinct wallets between October 30, 2025, and October 7, 2026. Bitquery found that 96% of detected attacks hit Pump.fun's launch curve or PumpSwap, the exchange associated with the memecoin platform. Wallets are addresses, not verified counts of individual people. Trades involving other bots accounted for 39% of estimated gross proceeds.
Bitquery counted 6,850,144 successful sandwich attacks on Solana from October 30, 2025, to October 7, 2026. The estimated $29.7 million represents gross proceeds before fees and other costs.
A bot buys before a user's swap, pushing up the pool price, then sells after the user's transaction. The user receives fewer tokens than they would have without the bot's trades, while the bot sells at the higher price. There may be no failed transaction or separate charge; the cost shows up in the execution price.
The distinction between an on-chain address and a person also matters in this tracing analysis. Blockchain records can show activity without independently proving identity or intent.
Slippage and exposure
On Jupiter-routed trades through Pump.fun's launch curve or PumpSwap, transactions with 20%-50% slippage tolerance were sandwiched 14 times as often as those below 3%, Bitquery found. Against trades below 0.5%, the observed rate was 96 times higher. Slippage tolerance sets the maximum price movement a trader will accept from the quoted rate. Jupiter applies that user-set limit to each submitted swap: a move beyond it can make the swap fail, while a wider limit can let it execute at a worse price.
Solana does not use Ethereum's traditional public-mempool model; transaction ordering depends on access to ordering within a slot and infrastructure such as validators, relays, and block builders. Bitquery also reported that most attacks identified in its September 2026 research no longer used the Jito-bundle path, so bundle-based protection does not eliminate the need for a low slippage limit.
The pattern did not rise steadily with each increase in tolerance. Trades below 0.5% were attacked more often than those set between 1% and 3%. The 50%-or-more band saw fewer attacks than the 20%-50% band. These figures describe a subset of routes; they do not prove that slippage settings alone caused the attacks.
Bitquery also found that the share of wallets allowing at least 10% slippage on half or more of their Jupiter trades climbed from 21% in November 2025 to 30% in September 2026, then 36% in the first week of October. The finding tracks wallet settings. It does not prove those settings caused any individual attack.
Jupiter accounted for only a small share of all victim trades in the investigation. The comparison covered only its routes through Pump.fun's launch curve or PumpSwap, so the rate differences should not be generalized to every Solana route.
Gross proceeds are not profit
Bitquery valued SOL at the minute of each attack to calculate its $29.7 million estimate. In a separate sample of 63 attacks that included fees, bots retained 68% of aggregate gross proceeds. Eighteen attacks lost money after costs. That small sample does not establish net earnings across the full 11-month period. Bitquery also flagged inconsistent recording of Pump.fun sale fees, which affects its gross estimate by about 3%.
The estimate has other limits. Bitquery identified bots through repeated trading patterns and excluded round trips that began within a token's first 10 slots. It also reported gaps in exchange and block coverage. Chain records show transaction order and outcomes, but cannot prove a wallet saw a user's order beforehand. Some wallets classified as bots may have been fast traders rather than sandwich attackers.
A separate research preprint submitted September 23, 2026, examined protected order flow across six blockchains. These are transactions intended to be shielded from front-running. The authors reported sandwich patterns extending across trades and blocks, and argued that protections may offer weaker guarantees than users expect. The work is a preprint, not a guarantee that every protected route is vulnerable.
Slippage controls set the price a trader will accept, but cannot by themselves stop another transaction from moving the pool first. Bitquery's $29.7 million figure measures gross activity under its method; it is not a verified tally of bot profits or individual losses. A wide tolerance gives a volatile swap more room to execute, while permitting a less favorable fill. The available data does not quantify risk across all Solana routes.
