Sandwich attacks on Solana have reached epidemic proportions, with 28 million recorded over three years. Yet traders using Jupiter Ultra appear to be catching significantly fewer of them, according to the first multi-year academic study of protected order flow attacks across major blockchains.

The research, titled “No Place to Hide: An Analysis on Protected Order Flow Sandwich Attacks,” was published on arXiv on September 23, 2026. Conducted by researchers from Category Labs, ETH Zurich, Flashbots, and the University of Lisbon, it examined sandwich attack activity across Ethereum, Solana, Tron, Base, Arbitrum, and Monad. The findings indicate that application-level design choices make an enormous difference in how often users get exploited. The “protected order flow” in the title refers to swap traffic submitted through private channels rather than public visibility — the same design principle that keeps transaction details concealed until execution.

How Sandwich Attacks Work

In a sandwich attack, a bot detects a pending swap, places a buy order immediately before it to push the price up, then sells right after the victim’s trade executes at the inflated price. The trader receives worse execution, and the bot pockets the difference. Because the cost shows up as a worse price rather than an explicit fee, it is easy to miss on any individual swap even as it adds up across the millions of attacks recorded during the study period.

Solana’s architecture makes the network particularly fertile ground for this type of exploitation. Low transaction fees allow bots to attempt sandwich attacks cheaply and at massive scale, while mempool visibility hands them the information needed to front-run trades.

Jupiter Ultra’s Numbers Versus the Competition

The study employed a metric called the “excess ratio” to measure how often each application’s users were victimized by sandwich attacks relative to a baseline. Jupiter Ultra recorded an excess ratio of 0.7 overall, meaning its users were actually sandwiched less often than statistical expectations would predict. In single-victim scenarios, the ratio rose to 2.0 — still well below the competition.

The contrast with other Solana trading terminals is stark. Axiom posted a ratio of 18.9, Photon hit 11.1, and both BullX and GMGN also exceeded ratios of 10 — figures indicating attack frequencies well above expected levels for their users.

Jupiter attributes the gap to Ultra V3, which launched on October 17, 2025, and introduced several protective features. Private routing conceals transaction details until execution, cutting off the information advantage that sandwich bots rely on. Dynamic slippage estimation adjusts tolerance levels in real time rather than relying on static, user-set parameters. The platform also incorporated what it calls “Iris meta-aggregation” and “Ultra Signaling” to further enhance transaction privacy and execution quality.

The claimed results are aggressive: sandwich protection 34 times better than competitors, average positive slippage of +0.6 basis points, and execution fees 8 to 10 times lower than rival platforms.

Architecture Matters More Than the Chain

Among the study’s more notable conclusions is that the underlying blockchain is not the only variable. Application-level decisions — how a trading platform routes orders, what it exposes to the public mempool, and how it handles slippage — play a decisive role in determining which users get attacked and how often. For traders, that means the terminal they route a swap through is itself a measurable factor in execution risk.

The researchers were nonetheless clear that sandwich attacks have not been eliminated on Solana. Validators still represent a vector of exposure, and while Jupiter Ultra reduces the problem, it does not solve it. The excess-ratio figures now give the industry a common yardstick for comparing platform-level protection — a benchmark that shows how wide the gap between applications currently runs, and how much of the attack surface remains outside any single platform’s control.