Are trading bots legit?The software is real. The promised profits usually aren’t.
A trading bot is a program that places orders by rules — buy at set intervals, buy low and sell high inside a range, quote both sides of a market, or follow a model’s signal. Exchanges offer such bots openly, so the technology is legit. Profit is a different matter: in the first public AI trading contest with real money, four of six models lost, and in a study of Brazilian day traders 97% of those who kept going for 300 days lost money. Meanwhile some of the largest crypto frauds sold a “bot” that never existed. Below: what bots do, what the evidence says, why costs eat returns, and how to check one.
Reviewed November 6, 2026 · figures from our live data
The short answer
Bots themselves are legit: exchanges such as Binance offer grid bots, and open-source tools run market-making and arbitrage strategies. A real bot trades on your own account or wallet.
Evidence on profits is weak. In Alpha Arena (October–November 2025, $10,000 each) four of six AI models lost money; the StockBench study found most models do not beat simply buying and holding.
The biggest “bot” losses were frauds: Mirror Trading International took at least 29,421 bitcoin and, the CFTC found, misappropriated all of it; the “bot” did not exist.
Red flags: guaranteed or fixed daily returns, money sent to the provider’s wallet, referral rewards, no licence. A real bot needs only a trade key that you can revoke — never your withdrawal rights.
tokens an agent’s own wallet can hold in our data today
issuers
stated yield range across 313 with a rate
What trading bots actually do
Most bots fall into a few families. None of them predicts the market; each makes money only in certain conditions.
- 1
Grid bots
A grid bot places buy and sell orders at preset intervals inside a price range. When a buy fills, it places a sell one level higher, and the other way round. Binance says the strategy suits volatile markets that move sideways within a range; if the price leaves the range, orders stop executing until it returns.
- 2
DCA bots
A dollar-cost-averaging bot buys a fixed amount at fixed times, regardless of price. It is automation, not a trading edge: the result is whatever the asset does over the period, minus fees.
- 3
Arbitrage bots
An arbitrage bot buys where a token is cheaper and sells where it is dearer. On blockchains this is a race: the 2019 paper “Flash Boys 2.0” documented bots bidding up transaction fees against each other to get their trades ordered first — the edge goes to the fastest and best-funded.
- 4
Market-making bots
A market maker posts buy and sell orders on both sides and earns the gap when both fill. Hummingbot, an open-source bot framework, names the main risk: in a trend one side keeps filling, and the bot builds up inventory of the asset that is losing value.
- 5
Signal and AI bots
These follow a trading signal — an indicator, a paid signal service, or a language model deciding what to buy. They are the hardest to judge: the quality of the signal is the whole product, and it is also what scams most often invent.
What the evidence says about profits
Independent evidence is thin, but it points the same way: running a bot is easy, beating the market with one is not.
- 1
Alpha Arena, season 1 (crypto)
From 17 October to 3 November 2025, Nof1 gave six AI models $10,000 each to trade crypto perpetuals on Hyperliquid on their own. Qwen3 Max ended +$2,232 and DeepSeek +$489; Claude Sonnet lost $3,081, Grok $4,531, Gemini $5,671 and GPT-5 $6,267, about 63%. Every model won only 25–30% of its trades.
- 2
Alpha Arena, season 1.5 (US stocks)
In the second round, which ended in early December 2025 and traded US stocks, a model named Grok 4.20 was up about 12% on average. On 8 December it was the only profitable model; the others were losing.
- 3
StockBench
A 2025 benchmark that let AI agents trade stocks over several months, with daily market data, found that “most models struggle to outperform the simple buy-and-hold baseline”. A few showed potential for higher returns and better risk control.
- 4
Human day traders as a benchmark
A study of everyone who started day trading Brazilian equity futures between 2013 and 2015 found that of those who persisted for at least 300 days, 97% lost money and only 0.4% earned more than a bank teller (US$54 a day). A bot following the same short-term strategies faces the same market.
- 5
Leveraged productsEU only
In the EU, ESMA cited analyses in March 2018 showing that 74–89% of retail CFD accounts typically lose money, and restricted CFD sales to retail clients. Many bots trade with leverage — perps, futures, CFDs — which magnifies these losses.
- 6
Backtests are not evidence
A 2014 paper by Bailey, Borwein, López de Prado and Zhu proved that impressive simulated performance is easy to reach after testing only a few strategy variants, and that such overfit strategies can have negative expected returns in live trading. A backtest that does not say how many variants were tried cannot be judged.
Why most retail bot strategies lose after costs
A bot does not remove the costs and competition that make short-term trading hard. It often adds to them by trading more.
- 1
Fees grow with every trade
In Alpha Arena, Qwen3 Max paid $1,654 in fees on a $10,000 account — most of a profit of $2,232. Gemini made 238 trades and paid $1,331 in fees while losing. A bot that trades often must earn its fees before it earns anything.
- 2
You compete with faster players
The Brazilian study concluded that it is virtually impossible for individuals to compete with high-frequency traders. On blockchains, arbitrage bots outbid each other on fees for priority; a retail bot on a home server arrives last.
- 3
Grid bots break in a trend
Binance warns that its grid bot keeps executing at set intervals even during a sharp crash or rapid rise, and recommends stop orders. In a falling market a grid keeps buying; in a rising one it sells too early.
- 4
Market makers hold the loser
Hummingbot describes inventory risk as the chance that a market maker cannot find buyers for what it holds. In a sustained trend the bot accumulates the asset that is losing value.
- 5
Strategies tuned to the past
A bot sold with a strong backtest proves little on its own. As the overfitting paper shows, the more settings someone tried, the more likely the best one only fits the past.
Checklist: a real bot or a scam?
A real bot is software that trades your money where you keep it. A scam asks you to hand the money over.
- 1
Where is the money?
A real bot trades on your exchange account or your own wallet; you can see every order there. If you must deposit into the provider’s wallet or “pool”, it is not a bot you control — Mirror Trading International, Trade Coin Club and the Adam brothers’ scheme all worked that way.
- 2
Which permissions does it need?
Trading needs a trade key, not withdrawals. Kraken lists “Withdraw Funds” as a separate permission and warns that a compromised key with it is a direct financial risk. Binance gives keys without an IP restriction read access only, and withdrawal permission requires one.
- 3
Can you lock and revoke the key?
Lock the key to the bot’s IP address — Kraken says this prevents almost all misuse even if the key leaks — and delete it in one click when you stop. In the 3Commas leak in late 2022, about 100,000 keys reportedly fell into attackers’ hands; HAPI verified $27.3 million in losses by January 2023, from trades into illiquid coins — no withdrawal right needed.
- 4
Guaranteed or fixed returns
Trade Coin Club promised at least 0.35% a day; the Adam brothers 13.5% a month; Privvy more than 100% in as little as 21 days. The CFTC calls promises of high or guaranteed returns a warning sign of fraud: no bot can promise what the market does.
- 5
Recruiting and referral levels
MTI and Trade Coin Club were run as multi-level marketing programmes; payouts came from new deposits. Rewards for bringing in new users are a sign of a scheme that needs new money.
- 6
Who runs it and is it licensed?
The CFTC advises checking the people behind a bot and how long its website’s domain has existed. In the UK, the FCA advises checking any firm on its Firm Checker; BaFin warned in 2025 about 17 bot sites with near-identical layouts and no registered office.
Cases: when the “bot” did not exist
Regulators have taken on bot schemes worth billions. In each case the claimed trading was the story, not the source of payouts.
- 1
Mirror Trading International (CFTC, 2023)
From about May 2018 to March 2021, MTI took at least 29,421 bitcoin from participants, including at least 23,000 in the US, claiming a proprietary forex “bot”. In April 2023 a US court ordered founder Cornelius Steynberg to pay $1.73 billion in restitution and an equal penalty — the largest penalty in any CFTC case.
- 2
Trade Coin Club (SEC, 2022)
The SEC alleged in November 2022 that Trade Coin Club raised more than 82,000 bitcoin, worth about $295 million, from over 100,000 investors from 2016 to 2018. It claimed a bot made “millions of microtransactions” every second; withdrawals came entirely from new deposits.
- 3
The Adam brothers (SEC, 2024)
In August 2024 the SEC charged Jonathan and Tanner Adam with raising $61.5 million from more than 80 people for an arbitrage bot that, the SEC says, never existed; it alleges $53.9 million was misspent, including on a $30 million condominium.
- 4
Privvy (SEC, 2026)
On 28 May 2026 the SEC sued Nathan Fuller of Texas, alleging he raised about $12.3 million from about 150 investors for “proprietary AI-based trading bots” doing arbitrage. It alleges $6.2 million went to personal expenses and about $5.5 million to Ponzi-like payments, with fake account statements.
- 5
AI washing (SEC, 2024)
In March 2024 the SEC fined two investment advisers, Delphia and Global Predictions, a combined $400,000 for claiming AI capabilities they did not have. Saying “AI” in the marketing is not proof that any AI is used.
What regulators say
Warnings from four continents describe the same pattern: AI and bots as a sales story for unrealistic returns.
- 1
United States: CFTC
In January 2024 the CFTC issued the advisory “AI Won’t Turn Trading Bots into Money Machines”: AI cannot predict the future or sudden market changes, and investors should factor in fees, spreads and subscription costs.
- 2
United States: FBI losses
The FBI’s Internet Crime Report for 2024 counted 149,686 complaints involving cryptocurrency with $9.3 billion in losses; crypto investment fraud alone accounted for $5.8 billion from 41,557 complaints.
- 3
Australia: ASIC
ASIC had 11,964 phishing and investment-scam websites removed in 2025, up 90% on the year before. Among the tactics it lists: trading bots that supposedly use AI to generate passive income. ASIC’s verdict: “It is very unlikely that these trading bots exist.”
- 4
United Kingdom: FCA
The FCA keeps a warning list of unauthorised firms; on 6 June 2025 it added “Ai Trader Bot”. Dealing with an unauthorised firm means no access to the Financial Ombudsman Service or FSCS protection.
- 5
Germany: BaFin
In February 2025 BaFin warned about 17 websites advertising an “AI-controlled algorithm” for trading crypto without authorisation. In January 2026 it published, with ESMA, EBA and EIOPA, fact sheets on AI and crypto fraud in all EU languages.
The 12 highest stated yields
Yields as stated by the issuer or read on chain, with the date they were checked on each token page. Not a recommendation.
| Token | Issuer | Yield | Minimum |
|---|---|---|---|
| Orca DEX — SOL / NEAR liquidity pool on Solana | Orca DEX (protocol) | 94.74% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Aerodrome Slipstream — WETH / USDC liquidity pool (CL50 - 0.025%) on Base | Aerodrome Slipstream (protocol) | 80.25% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Orca DEX — SOL / HYPE liquidity pool on Solana | Orca DEX (protocol) | 72.31% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Uniswap V3 — USDC / WETH liquidity pool (0.3%) on OP Mainnet | Uniswap V3 (protocol) | 64.95% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Raydium AMM — COINX / USDC liquidity pool (Concentrated - 0.8%) on Solana | Raydium AMM (protocol) | 63.7% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Uniswap V3 — WETH / USDC liquidity pool (0.3%) on Base | Uniswap V3 (protocol) | 59.88% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Raydium AMM — WSOL / ZEREBRO liquidity pool (Standard - 0.25%) on Solana | Raydium AMM (protocol) | 55.9% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Uniswap V3 — WETH / USDT liquidity pool (0.3%) on Ethereum | Uniswap V3 (protocol) | 53.25% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Uniswap V4 — UNI / USDC liquidity pool (0.30%) on Ethereum | Uniswap V4 (protocol) | 52.91% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Orca DEX — SOL / USDC liquidity pool on Solana | Orca DEX (protocol) | 52.55% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Uniswap V4 — ETH / USDT liquidity pool (0.30%) on Ethereum | Uniswap V4 (protocol) | 51.63% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
| Uniswap V3 — CLANKER / WETH liquidity pool (1%) on Base | Uniswap V3 (protocol) | 50.94% (30-day average of trading fees, DefiLlama 2026-11-11) | None |
Questions
Are trading bots legit?+
The technology is: exchanges and open-source tools run grid, DCA, arbitrage and market-making bots every day. What is often not legit is the offer around it — a “bot” that requires a deposit into someone else’s wallet and promises fixed returns.
Are trading bots profitable?+
Sometimes, for some strategies in some markets — but there is no public evidence that retail bots are profitable on average. In Alpha Arena four of six AI models lost money; StockBench found most models do not beat buy-and-hold, and fees eat into frequent trading.
Do AI trading bots work better?+
Not reliably. The 2025 contests showed large differences between models and between rounds; in the stock round only one model was in profit. The CFTC’s summary: AI cannot predict the future or sudden market changes.
How do I know a trading bot is a scam?+
Guaranteed or fixed daily returns, a deposit into the provider’s wallet, rewards for recruiting, no named people or licence, and a website whose domain is only months old. Any one of these is reason to stop.
Is it safe to give a bot my exchange API key?+
Only a key without withdrawal rights, locked to the bot’s IP address, on an account holding money you can afford to lose. The 3Commas leak showed that even trade-only keys can be used to drain an account through trades into illiquid coins — delete keys you no longer use.
I lost money to a bot scheme. What now?+
Stop paying any “fees” to withdraw, keep all records, and report it to your regulator or the police — in the US the FBI’s IC3, in the UK the FCA, in Australia ASIC. The CFTC warns that victims may not get money back, because the fraudsters may lack the funds.
- Are trading bots legal? →
- Trading bot vs human: who trades better? →
- Can AI agents trade on their own? →
- Build your own AI trading agent: step-by-step tutorial →
- What happens to your tokens if an exchange is hacked →
- RWA risks explained →
- Binance — What is spot grid trading
- Daian et al. — Flash Boys 2.0 (arXiv 1904.05234)
- Hummingbot — What is inventory risk
- Bitcoin.com — Alpha Arena launch on Hyperliquid
- Protos — Alpha Arena season 1 results
- ForkLog — Alpha Arena season 1.5
- StockBench (arXiv 2510.02209)
- Chague, De-Losso, Giovannetti — Day trading for a living?
- Bailey et al. — Pseudo-Mathematics and Financial Charlatanism (2014)
- ESMA — Restrictions on CFDs and binary options (Mar 2018)
- Kraken — API keys
- Binance — How to create API keys
- Cointelegraph — 3Commas confirms API key leak
- ForkLog — HAPI analysis of the 3Commas incident
- CFTC — Steynberg / Mirror Trading International judgment (8696-23)
- SEC — Trade Coin Club (LR-25571)
- Cointelegraph — SEC charges Adam brothers over $60M bot Ponzi
- SEC — Nathan Fuller / Privvy (LR-26558)
- SEC — AI washing charges (Mar 2024)
- CFTC — AI won’t turn trading bots into money machines (8854-24)
- FBI IC3 — 2024 Internet Crime Report
- ASIC — 26-063MR AI-powered investment scams
- FCA — Ai Trader Bot warning
- BaFin — Warning on AI trading-bot websites (Feb 2025)
- BaFin — AI and crypto fraud warning (Jan 2026)
This guide explains how these products work and is not investment advice or an offer of any security. Who may buy a product is set by its issuer; check the offering documents before investing.