A retail trader on Solana observes a promising new meme coin launching on Pump.fun and decides to buy 10 SOL worth during the early bonding curve phase. They submit their transaction to the mempool with what appears to be a reasonable fee, expecting execution within the next block or two. Instead, their transaction is preceded by an identical purchase order that was not visible before, and followed immediately by a large sell order that dumps the token’s price. By the time their transaction settles, the token price is higher than anticipated, they receive fewer tokens, and a bot has profited from the difference. This is a sandwich attack, and it has become a systemic feature of Pump.fun trading rather than an exception.
The problem stems from the intersection of three factors: Solana’s transparent mempool architecture, Pump.fun’s bonding curve mechanics, and the economic incentive for maximal extractable value (MEV). Unlike Ethereum’s encrypted mempools or protocols with threshold encryption, Solana nodes broadcast pending transactions publicly before they are confirmed. Market makers, bots, and searchers can observe the mempool, identify profitable opportunities, and construct multi-transaction bundles that extract value from ordinary traders. Pump.fun’s design makes this particularly lucrative because bonding curves create predictable price movements: if a large buy order is coming, a bot can buy first, let the retail order execute at a worse price, then sell immediately after. The retail trader pays the difference, and the bot captures it. Understanding how these attacks work, where the vulnerabilities lie, and what defenses exist is essential for anyone trading on the platform.
How Pump.fun’s bonding curve creates exploitable price movement
Pump.fun tokens use an automated market maker (AMM) model called a bonding curve, which ties token price directly to supply. As more SOL is deposited, the token supply increases and price rises according to a mathematical formula. When tokens are sold, the opposite happens: supply decreases and price falls. This model is deterministic and publicly visible—anyone can calculate the exact price impact of a given transaction size before it executes. A bot observing the mempool can therefore see an incoming 10 SOL purchase order and calculate that it will push the token price up by a specific percentage, say 8 percent. That same bot can then sandwich the trade: buy first at the lower price, watch the retail order execute at the higher price, and sell immediately into the elevated price created by that retail order.
The mathematical predictability of a bonding curve is its core strength as a mechanism for fair token launches—it eliminates insider allocation and presale manipulation. But that same predictability becomes a weakness in an environment where transaction ordering is observable and modifiable. Unlike a traditional order book where buyers and sellers must be matched and where large orders can be split or routed privately, a bonding curve allows anyone with transaction visibility to construct a risk-free trade: frontrun, let the larger order move the price, backrun. The sandwich attacker faces no inventory risk because they are not betting on price direction. They are exploiting the mechanical price movement created by the victim’s own transaction.
The bonding curve on Pump.fun continues until a “graduation” threshold is reached—typically when a certain amount of SOL has been raised or the token’s market cap reaches a target. Once graduated, the token transitions to a decentralized liquidity pool, usually on Jupiter or Raydium, where the dynamics change. Before graduation, however, the bonding curve is the sole price discovery mechanism, and it is transparent to all parties. This transparency, while essential for a no-code launch platform serving millions of users, creates the sandwich vulnerability. A searcher with access to Solana’s mempool and the computational resources to construct and submit bundles can see every pending order and extract value from it.
Solana’s mempool architecture enables MEV extraction at scale
Solana’s validator network operates with a transparent mempool by default. Transactions submitted to the network are broadcast to all nodes and become visible to anyone monitoring the network before they are included in a block. This design contrasts with Ethereum after the merge, where the mempool is encrypted or obscured in various ways to prevent front-running. Solana chose transparency partly for efficiency—the network’s single leader validator architecture relies on rapid transaction propagation and execution—and partly for operational simplicity. Validators do not need to decrypt or verify a hidden transaction state; they can see all orders and execute them in a straightforward order.
The cost of this transparency is exposure to maximal extractable value (MEV) extraction. A bot operator can rent access to a validator’s transaction stream, observe pending orders, and pay the validator a portion of extracted value in exchange for priority. Alternatively, a bot can submit its own transactions with higher fees to gain priority in block construction. Some searchers operate their own infrastructure or partner directly with builders and validators. The economic incentive is substantial: on Pump.fun alone, sandwich attacks have generated millions of dollars in MEV. Major searchers and hedge funds have deployed automated systems to detect and exploit bonding curve trades in real time.
It is important to note that MEV extraction on Solana is not technically a bug or protocol violation—it is a feature of the network’s transaction model. Solana’s validators have full visibility into pending transactions and control the order in which they are included. Extracting value from that ordering power is rational behavior given the incentive structure. The issue is that retail traders are not aware of this dynamic or do not fully price in the cost. A trader buying on Pump.fun believes they are getting the bonding curve price, but they may be getting a price that has already been inflated by a bot’s frontrun transaction.
The mechanics of a typical Pump.fun sandwich attack
A sandwich attack on a pump fun bonding curve typically unfolds in five steps. First, a bot monitors the Solana mempool in real time, watching for transactions that interact with specific Pump.fun token accounts. Second, when the bot observes a large buy transaction, it extracts key details: the account address, token mint, expected size, and fee level. Third, the bot constructs a frontrun transaction—an identical or slightly larger buy on the same bonding curve, with a higher fee to ensure it lands before the victim’s transaction.
Fourth, the bot bundles its frontrun transaction, the victim’s transaction, and a backrun sell transaction into a single Solana transaction bundle. This bundle is submitted to validators or builders with instructions that all three transactions should execute in sequence within a single slot. Fifth, the bundle is executed: the bot’s buy pushes the price up, the victim’s buy executes at this higher price, and the bot’s sell extracts the profit. The retail trader ends up with fewer tokens than expected, and the bot’s profit is the difference between what the victim paid per token and what the bot sold at.
The attack is nearly risk-free for the attacker because the sequence is atomic. If the victim’s transaction is large enough, the bot’s profit is guaranteed by the mathematics of the bonding curve. The attacker does not need to guess whether the market will move in their favor; they are extracting value from the mechanical operation of the bonding curve itself. The victim, by contrast, enters the transaction believing they are trading against a deterministic pricing curve, unaware that their own order has been reordered and bookended by extraction.
From the victim’s perspective, the attack is nearly invisible. The transaction appears to succeed—tokens are received, SOL is debited—but at a worse effective price than was advertised on the interface. No hacker accessed their wallet, no funds were stolen in the traditional sense. Instead, the sandwich attacker exploited the ordering of transactions to extract value that rightfully belonged to the victim as MEV. The attacker’s profit comes directly from the victim’s slippage.
Why retail traders on Pump.fun are particularly vulnerable
Pump.fun has democratized token creation and trading but at the cost of moving unsophisticated traders into an environment where MEV extraction is both common and technically sophisticated. Most users trading on the platform are not blockchain engineers; they are retail speculators looking to buy early-stage meme coins. They may not understand bonding curves, let alone concepts like transaction ordering, mempool visibility, or sandwich attacks. The interface displays a price and an estimated output, and traders naturally assume that submitting a transaction will execute at or near that displayed price.
This gap between perceived and actual execution is the core vulnerability. A retail trader does not have access to the mempool before their transaction lands; they cannot see the bot’s frontrun transaction preparing to execute. They cannot adjust their transaction fee in real time based on competitive conditions in the next block. They submit their order and wait, with no visibility into the ordering dynamics that will determine their actual execution price. By the time the transaction settles and they can see the result, the sandwich is complete and the damage is done.
The problem is compounded by the speed of Solana’s network. Blocks are produced every 400 milliseconds, and transactions propagate through the network in tens of milliseconds. A sandwich attack can be detected, bundled, and executed faster than a human trader can react. Even a moderately savvy trader who checks the mempool cannot outrun an automated bot. The asymmetry in speed and information favors the bot operator, not the retail trader.
Additionally, many retail traders are drawn to Pump.fun precisely because of its low barriers to entry and high potential returns. This same population tends to be less familiar with advanced trading concepts and more likely to use default settings. A bot operator can target these traders specifically, knowing that they are less likely to use protective measures or understand the cost of MEV. The result is a wealth transfer from retail traders to sophisticated searchers, mediated by the transparency of Solana’s mempool and the mathematical predictability of bonding curves.
Detection and quantification of sandwich attacks
Sandwich attacks on Pump.fun can be detected by analyzing on-chain transaction history and clustering trades by time, value, and token account. Researchers have built tools that identify probable sandwich patterns: a large buy, preceded by a smaller buy from the same wallet or related addresses, followed by a corresponding sell. The temporal and value clustering of these patterns is rarely random. Statistical analysis of attack frequency suggests that a significant percentage of large trades on Pump.fun experience some degree of sandwich extraction.
Quantifying the total cost to retail traders is more difficult because it requires counterfactual analysis—calculating what price the victim would have received in the absence of the sandwich. Some analyses use the stated bonding curve formula to compute the theoretical price for a given transaction size, then compare it to the actual price received. Others look at the difference between the displayed price at the time of transaction submission and the effective price paid per token after settlement. These methods suggest that sandwich costs can range from a fraction of a percent on small trades to 10–20 percent on larger orders, depending on competitive conditions and bot strategies.
The total MEV extracted from Pump.fun is substantial. Conservative estimates suggest hundreds of thousands of dollars per day in aggregate sandwich attacks across all tokens on the platform. On volatile tokens with large volume, the figure can be much higher. Some individual attacks have extracted six figures from a single trade. For comparison, the retail traders themselves are collectively making speculative bets on meme coins that may or may not appreciate; the sandwich MEV is pure value extraction independent of price movement. From an economic standpoint, MEV is a regressive tax that hits retail traders hardest because they have the least visibility and least ability to protect themselves.
Existing and potential defenses against sandwich attacks
Several defenses have been proposed or partially implemented, though none completely eliminates the risk. The first is slippage protection, where a trader specifies a minimum token output below which the transaction will revert. Pump.fun’s interface allows users to set slippage tolerance, but many retail traders either do not understand this feature or leave it at the default 50 percent, which offers minimal protection. A sandwich attacker can often make a profit even if the victim’s slippage limit is moderate, by carefully calibrating the size of the frontrun and backrun.
A second defense is MEV-resistant routing through private mempools or encrypted transaction pools. Some platforms like Jito and MEV-blockers offer private transaction ordering to prevent frontrunning. However, these solutions are not widely integrated into Pump.fun’s interface, and they require additional trust in a third party (the MEV solution provider) instead of removing MEV entirely. They are also less effective on Solana because the network’s architecture gives validators strong ordering power regardless of where a transaction originates.
A third potential defense is threshold encryption or encrypted mempools
A fourth defense is application-level ordering where the bonding curve itself includes mechanisms to batch trades or execute them in a predetermined order rather than immediately. Some newer token launch platforms have experimented with batch auctions or commit-reveal schemes that reduce MEV extraction. However, these mechanisms typically reduce liquidity and increase latency, trading off execution speed for extraction resistance. Pump.fun’s core appeal is instant trading with immediate price discovery, so such mechanisms may be incompatible with its design philosophy. The fifth and most practical defense for retail traders is awareness and conservative trading practices. Users should understand that sandwich attacks exist, set reasonable slippage limits, avoid trading extremely large amounts on bonding curves, and be skeptical if their effective execution price differs significantly from what they expected. Some traders use tools to monitor their transactions post-execution and flag unusual MEV extraction. However, this places the burden entirely on retail users to educate themselves and adopt protective behaviors, rather than addressing the structural vulnerability. Addressing sandwich attacks on Pump.fun would require changes at multiple levels: protocol changes to Solana’s mempool, application-level changes to how bonding curves price transactions, and behavioral changes from users. Each introduces trade-offs that Solana’s ecosystem has been unwilling to accept. Protocol-level changes like threshold encryption would slow block production and reduce throughput, undermining Solana’s core value proposition as a high-speed chain. Application-level changes like batch auctions would make Pump.fun slower and less responsive, potentially reducing its appeal to users seeking instant gratification. There is also limited incentive for Solana validators to prevent MEV extraction. Validators benefit from MEV through direct extraction or delegation to searchers. Solana Foundation and the broader ecosystem have focused more on scaling throughput and reducing fees than on MEV mitigation. This contrasts with Ethereum, where MEV has become a more prominent policy concern. As Ethereum has explored encrypted mempools and threshold encryption, Solana has largely accepted MEV as a feature of the network. For Pump.fun specifically, the platform’s business model does not depend on preventing MEV. The platform takes a 1 percent fee on trades, and this fee is unaffected by whether the trade is sandwiched. Pump.fun has no direct incentive to eliminate sandwich attacks, though reducing them might improve user experience and retention. Some third-party solutions will likely emerge, and more sophisticated traders will adopt protective measures. But for the bulk of retail traders using the default interface, sandwich exposure will likely remain a persistent cost of trading on Pump.fun and other Solana DEXs. The Pump.fun sandwich problem is not unique to that platform—it is symptomatic of a broader tension in decentralized finance on transparent blockchains like Solana. Any DEX or bonding curve system that operates with public transaction visibility and deterministic pricing mechanics will face MEV extraction. Raydium, Jupiter, and other Solana DEXs experience similar attacks. The problem extends to newer chains like Aptos and Sui that have adopted similar mempool transparency models. The challenge is that the features enabling MEV extraction—transaction visibility, deterministic pricing, fast block times—are also the features that make these systems attractive for certain use cases. Solana’s speed and low fees draw retail users and developers. Bonding curves’ determinism and fairness (compared to presales) draw token creators. Trading on Pump.fun at low friction is valuable precisely because the platform is transparent and immediate. Completely eliminating MEV would require sacrificing these properties. The more realistic outcome is an ecosystem with different trade-offs. Platforms optimizing for MEV resistance (like privacy-focused DEXs or those using encrypted transactions) will be slower and more expensive. Platforms optimizing for speed and accessibility (like Pump.fun) will expose users to sandwich attacks. Sophisticated traders will use tools and protective measures to reduce MEV exposure. Retail traders will continue to subsidize bot operators, either by accepting the cost or by learning to navigate the landscape. The pump fun platform has made token creation democratic, but it has also made MEV extraction democratic in turn—available to anyone with the technical sophistication to run a bot and the capital to pay for transaction ordering. The cost varies depending on trade size, token volatility, and competitive conditions. Small trades might lose 1–5 percent to sandwich MEV, while larger trades can lose 10–20 percent or more. The cost is the difference between the bonding curve price the trader expected and the actual effective price they received after the frontrun bot’s transaction moved the price upward. Yes, partially. Set a strict slippage limit (5–10 percent rather than 50 percent), avoid trading extremely large amounts that are particularly attractive to sandwich bots, and monitor your actual execution price post-transaction. However, there is no perfect protection at the application level on Solana. More sophisticated defenses like private mempools exist but require additional trust and complexity. Fixing it would require protocol-level changes (like encrypted mempools) that would slow Solana’s block production, or application-level changes (like batch auctions) that would reduce the instant-trade functionality that makes Pump.fun attractive. Additionally, validators and searchers benefit from MEV extraction, creating no economic incentive to eliminate it. The problem is structural rather than a bug that can be patched.Why the problem is unlikely to be solved in the short term
The broader implications for decentralized finance on transparent blockchains
Frequently asked questions
How much does a sandwich attack typically cost a retail trader on Pump.fun?
Can I protect myself from sandwich attacks when trading on Pump.fun?
Why hasn’t Solana or Pump.fun fixed the sandwich attack problem?