Research
From Hype to Collapse: Investigating Rug Pull Scams on Solana
Rug Pulls — where token creators attract buyers and then drain value through freezing accounts, pulling liquidity, or mass-selling — are rampant on Solana, with prior work cited in the paper…

- arXiv
- 2603.24625
- Published
- 2026-03-25
- Authors
- Jiaxin Chen, Ziwei Li, Zigui Jiang, Ruihong He, Yantong Zhou, Jiajing Wu, Zibin Zheng
AI summary
Overview
Research area: Blockchain security and Web3 measurement — specifically fraudulent token schemes (Rug Pulls) on the Solana blockchain, analyzed through on-chain transaction and state data.
Technical level: Intermediate. The paper is readable without deep blockchain expertise, but familiarity with DEXs, liquidity pools, and token minting helps.
Scope (one sentence): A large-scale empirical measurement of Solana Rug Pulls that builds the first public Solana Rug Pull dataset (a 117-token manually verified benchmark plus 76,469 pipeline-labeled in-the-wild tokens) and characterizes the ecosystem along naming, off-chain propagation, lifecycle, economics, and syndicate organization.
What This Paper Is About
Rug Pulls — where token creators attract buyers and then drain value through freezing accounts, pulling liquidity, or mass-selling — are rampant on Solana, with prior work cited in the paper estimating that roughly 55% of newly issued Solana tokens or liquidity pools eventually show Rug Pull behavior. Unlike Ethereum, Solana tokens all run on the shared SPL Token program, so scammers cannot embed malicious logic in a custom contract; fraud instead happens through on-chain behavior, which existing Ethereum-centric detection research does not address. The authors' goal is to establish the first systematic, Solana-specific empirical foundation: define the behavioral patterns, build and release a large labeled dataset, and measure what the scam ecosystem actually looks like.
Key Contributions
-
Behavioral taxonomy and verified benchmark. The authors manually verify 68 community fraud reports (from Chainabuse, plus analysis from SlowMist and CertiK) into a benchmark of 117 confirmed Rug Pull tokens, and distill three representative on-chain execution patterns: Freeze Authority Abuse, Liquidity Withdrawal, and Pump-and-Dump. Every token in the benchmark maps unambiguously to one pattern (2 / 26 / 89 respectively).
-
A behavior-guided identification and human-validation pipeline that works only on raw on-chain transaction and state data — deliberately avoiding aggregated price feeds from GMGN, Birdeye, DexTools or DexScreener, which the authors note disagree on price, percentage change and FDV for the same token at the same timestamp. Applied to 100,063 tokens newly issued on Orca, Raydium and Meteora between January 1 and June 30, 2025, it yields 76,469 labeled Rug Pull tokens.
-
First publicly released large-scale Solana Rug Pull dataset, comprising Benchmark-117 (manual ground truth) and ITW-76469 (in-the-wild pipeline labels: 60,402 Pump-and-Dump, 15,606 Liquidity Withdrawal, 461 Freeze Authority Abuse). Label reliability is estimated by a stratified random audit of 382 tokens (~0.5% of candidates) at a 0.26% false-positive rate (95% upper bound 1.45%).
-
A five-dimension ecosystem measurement — token naming, off-chain propagation, lifecycle dynamics, economic impact, and syndicate organization — quantifying at least $151M in direct losses across 7,322 profiting addresses and identifying 78 large-scale syndicates organized into two recurring on-chain topologies named Star and Cluster.
Main Findings
-
Rug Pull tokens die almost immediately. Median lifespan is 0.0116 days versus 375.4549 days for legitimate tokens (means: 1.0339 vs 416.5262). Median holder count is 9 versus 41,026, and median transaction rate is 0.08 per hour versus 299.11. The median share of DeFi activity occurring on day one is 100% for Rug Pull tokens versus 0.00% for legitimate ones.
-
Naming is used as camouflage. Rug Pull token names cluster around trending themes — "AI", "agent", "trump" — alongside persistent pet memes like "doge" and "pepe". Two deceptive patterns are quantified: 3,203 tokens with deliberately inconsistent name–symbol pairs (symbols mimicking USDT, SOLANA, X, SORA, etc., which matters because many DEXs and wallets display the symbol), and 1,030 look-alike tokens imitating known brands through direct imitation, suffix manipulation, or case confusion (e.g., USDTea, USDTf, TIKTOK).
-
Off-chain presence is thin and disposable. Of the 76,469 fraudulent tokens, only 19.0% (14,552) disclose an X account and 14.3% (10,957) provide a website. Of those website links, only 29.3% are authentic functional project sites; over 70% are invalid or replaced by social-media homepages and irrelevant sites. Of the X links, over 60% do not correspond to normally operating accounts — 9.9% point to celebrity or trending accounts, 30.6% to single-use promotional posts, and the remainder to banned or deactivated accounts — with only 19.9% showing basic project-level maintenance.
-
Freeze Authority is a Solana-specific trap. Because all Solana tokens run on SPL Token, a creator who retains the freeze authority can invoke the
FreezeAccountinstruction so victims can buy but cannot sell, while insider accounts sell freely. This category is real but comparatively rare (461 of 76,469 pipeline labels; 2 of 117 in the benchmark). -
Pump-and-Dump dominates by volume. 60,402 of 76,469 labeled tokens, and 89 of 117 benchmark tokens, use this pattern, in which scammers sometimes voluntarily renounce LP tokens or freeze authority to appear "fair-launched" and then sell pre-allocated holdings, sometimes dispersed across many addresses to evade monitoring.
-
Pump-and-Dump detection is tuned conservatively, producing a lower bound. The holder-decline threshold τ_down was swept from 0.50 to 1.00 in 0.01 increments and set to 0.73 — the largest value that still admits gradual sell-offs. The authors explicitly treat labels as a lower bound: in a check of 100 tokens labeled non-Rug Pull, 9 were missed cases, all unfolding over three to seven days (outside the 24-hour observation window).
-
Single false positive illustrates a real edge case. The one error found in the 382-token audit was "The First Revenue Backed Community Coin", which showed Rug Pull-like behavior early but was later revived through a community takeover.
-
Losses and organization are substantial. The paper reports at least $151M in direct losses, 7,322 profiting addresses, and 78 syndicates, which it says form a maturing fraud supply chain on Solana; the abstract also characterizes lifecycles as tightly coupled to the SOL price. (Detailed RQ2/RQ3 analysis sections fall beyond the truncated text provided; the summary figures above come from the abstract and contribution list.)
Methodology in Plain English
The authors worked in two stages. First, they built a small, trustworthy reference set: they searched Chainabuse and security-firm write-ups for Solana Rug Pull reports, verified 68 of them, and extracted 117 fraudulent tokens. Three researchers with 2+ years of blockchain anti-fraud experience sorted these using an open card-sorting approach, cross-checking on-chain platforms (Solscan, GMGN, Dextools, Birdeye, Dexscreener) until they agreed on both the fraud type and execution pattern. Comparing these 117 tokens against 65 verified legitimate tokens (from Jupiter's isVerified mechanism with an Organic Score of at least 80) produced the statistical signature that drives the rest of the work: short life, few holders, liquidity exhaustion, and concentrated day-one activity.
Second, they scaled up. They pulled every newly issued token on Orca, Raydium and Meteora — which together account for over 80% of Solana trading volume — for the first half of 2025 via the official Solana RPC and Solscan API, keeping raw transactions and state events rather than third-party price aggregations, so any label can be reconstructed from primary sources. A loose activity filter drops tokens still averaging 5+ transactions per hour over the trailing 24 hours (results were stable across 0.5–5). The survivors are tested against the three behavioral patterns: an actually-executed FreezeAccount against a user while the creator still holds freeze authority; a profitable liquidity removal by the mint authority or initial pool deployer followed by collapse; or monotonic decline in both holders and pool balance beyond the τ_down threshold. Because precision was prioritized over coverage, a random 0.5% sample (382 tokens) was stratified by pattern and independently audited by three researchers under a structured protocol, with disagreements resolved by discussion. Notably, the authors deliberately exclude Pump.fun-only tokens that never establish a liquidity pool, on the grounds that a cited report found up to 98.6% of Pump.fun tokens become worthless pump-and-dumps, making them low-value analytical noise.
Why This Matters
Impact on research: Prior Rug Pull research is largely EVM-centric — it leans on contract code and hidden functions as detection signals, which simply do not exist on Solana, where every token is governed by the same SPL Token program. This paper redirects the field toward behavior-driven detection and supplies the first public Solana Rug Pull dataset, giving other researchers a shared benchmark where none previously existed.
Real-world applications:
- Wallet and DEX risk warnings — flags for retained freeze authority, creator-side liquidity removal, and the 24-hour holder-decline cliff could be surfaced at the point of trade.
- Naming-based scam filters — the 3,203 name/symbol mismatches and 1,030 look-alike tokens give concrete heuristics for interfaces that show symbols prominently.
- Deprioritizing social signals in risk scoring — with only 19.0% of scam tokens linking an X account and 14.3% a website, and most of those links invalid or single-use, off-chain presence is a weak trust proxy.
- Law-enforcement and exchange tracing — the identification of 7,322 profiting addresses and 78 syndicates with Star and Cluster topologies gives investigators concrete network structures to target rather than isolated tokens.
Industry relevance: The paper quantifies a mature fraud supply chain rather than isolated opportunism, which matters for launchpad operators (Pump.fun-style one-click issuance), DEX platform policy, LP-token custody norms, and any compliance team building crypto risk models. The authors also point out that the gap in Solana-specific understanding has left stakeholders without a reliable basis for calibrating risk-assessment services, timely warnings, or mitigation strategy.
Future Directions
- Detecting slow rugs. The pipeline's 24-hour observation window systematically misses variants that unfold over three to seven days — the 9 of 100 missed cases in the audit were all in this category. Extending the window without flooding the dataset with noise is an open design problem.
- Evaluating "community takeover" revivals. The single audit false positive was a token that looked fraudulent early but was later revived by its community. Separating genuine revivals from staged ones is unresolved and directly affects label quality at scale.
- Turning patterns into deployable detectors. The paper motivates on-chain defense mechanisms but the three behavioral definitions are described as a labeling pipeline, not as a real-time, low-latency detector; measuring precision and latency in a live setting is a natural next step.
- Characterizing the Star and Cluster syndicates in depth. The two recurring topologies and the 78 identified groups suggest follow-on work on how these organizations recruit, share infrastructure, and rotate funds — and whether disrupting the topologies is more effective than blacklisting individual tokens.
Target Audience
Blockchain security researchers and graduate students working on fraud detection and measurement studies; DEX, wallet, and launchpad security engineers building rug-pull warnings; crypto compliance and threat-intelligence analysts tracing scam networks; and regulators or ecosystem stakeholders who need an empirical — rather than anecdotal — picture of how token fraud operates on Solana. Readers with basic familiarity with liquidity pools and token minting will get the most out of it.
Authors’ abstract
Solana has experienced rapid growth due to its high performance and low transaction costs, but the extremely low barrier to token issuance has also enabled widespread Rug Pulls. Unlike Ethereum-based Rug Pulls, which often rely on malicious smart-contract logic, Solana's unified SPL Token program shifts fraudulent execution toward on-chain behavioral manipulation. However, existing research has not systematically examined these Solana-specific Rug Pull patterns, and no…
The full abstract on arXiv