11 AI Trading Bot Claims You Should Verify Before Depositing $1
A polished dashboard, an AI logo and a screenshot showing 95% wins are not evidence that a trading system works. Before sending even one dollar, verify what is actually AI, who controls the funds, whether performance happened live, what permissions the software receives, what it really costs and whether withdrawals work.
What Matters
The most dangerous AI trading claim is not necessarily a lie. It is a claim that sounds precise while leaving out the information needed to verify it.
- A 95% win rate tells you almost nothing without average win, average loss, trade count, period, fees and live-order evidence.
- Guaranteed or unusually high returns are a major red flag. The CFTC specifically warns that AI cannot predict sudden market changes and that claims of guaranteed or spectacular returns are commonly used in fraud.
- Backtests are evidence about the past, not proof of future profitability. They can omit slippage, use favorable parameters or reflect market conditions that may never repeat.
- Demo profits are not live profits. Paper environments may behave differently because of liquidity, slippage, latency and fill assumptions.
- “Audited” is meaningless unless you can identify who performed the audit, what was audited, which accounts were covered and over what period.
- “AI-powered” does not tell you what the AI does. It may mean an autonomous agent, machine learning, parameter optimization, natural-language translation or ordinary algorithmic automation.
- “Non-custodial” does not automatically mean low risk. A trading credential that cannot withdraw funds may still have enough authority to create large losses.
- “Safe API connection” needs a permission map. Check Read, Trade, Derivatives, Transfer, Withdrawal, IP restrictions, expiry and revocation separately.
- Free software can still create expensive trading. Exchange fees, spreads, slippage, funding, model/API charges and turnover matter.
- Never accept “regulated” at face value. Verify the exact legal entity, regulator, registration number and scope of authorization independently.
- Never assume displayed profits are withdrawable profits. Test the withdrawal mechanism before materially increasing capital.
DN Evidence Block
- Verification date: 8 October 2026.
- Regulatory evidence: CFTC, SEC, Investor.gov and FINRA/NFA registration-verification resources.
- Recent enforcement context: September 2026 SEC and CFTC complaints involving alleged false AI, algorithmic-trading, regulatory and performance claims.
- Product evidence: current Coinrule, 3Commas and Cryptohopper documentation used to demonstrate how legitimate products disclose demo limitations, permissions and the actual role of AI.
- Core standard: marketing claim → independently verifiable evidence → explicit limitations.
- Not sufficient evidence: screenshots, testimonials, social-media posts, account dashboards, app-store availability, influencer endorsements or unaudited backtests on their own.
- DN testing status: this is a due-diligence framework, not a certification of any platform.
On September 29, 2026, the SEC announced charges against entities behind two alleged investment-confidence schemes. According to the complaints, supposed AI-generated signals, AI trading bots, fictitious profits and false SEC-compliance claims were used to attract investors. In one alleged scheme, investors attempting withdrawals were told they needed to pay additional advance fees. The SEC alleges there was no genuine trading platform and no actual AI-bot trading.
Four days earlier, the CFTC charged Cash FX Group and others in an alleged $950 million fraud. The complaint alleges investors were told expert traders, proprietary algorithms and artificial intelligence could generate returns as high as 15% per week, while minimal forex trading actually occurred.
The Rule: Extraordinary Automation Claims Need Extraordinary Verification
Trading software has always been difficult to evaluate.
AI makes the problem harder.
A traditional bot might say:
“Buy when indicator X crosses indicator Y.”
You can inspect the rule.
An AI product can instead say:
“Our proprietary neural intelligence continuously adapts to changing markets.”
That sounds more sophisticated.
It is also harder to falsify.
A vague technological claim can become a shield around an ordinary trading strategy, a weak model or, in the worst case, a product that does not exist.
The DN Claim-to-Evidence Rule
DN uses a simple principle:
The stronger the claim, the stronger the evidence required.
If a company says:
“We support paper trading”
you can usually verify that by opening the product.
If it says:
“Our AI delivers a 95% win rate”
a screenshot is not sufficient evidence.
The verification burden rises with the magnitude and financial consequence of the claim.
DN calls this:
Verification Burden Scaling.
A financial claim should require stronger evidence as its promised certainty, return or economic consequence increases. “We have an AI assistant” requires product evidence. “Our AI reliably generates exceptional returns” requires substantially stronger performance evidence.
The 11 Claims to Verify
“Our Bot Has a 90%, 95% or 100% Win Rate”
Win rate is probably the most seductive trading-bot statistic because it looks simple.
It can also be almost meaningless.
A system could win 95 trades by earning $1 each and lose five trades by losing $100 each.
The win rate would be:
95%.
The strategy would still lose money.
Ask for:
- total number of trades,
- average winning trade,
- average losing trade,
- profit factor,
- maximum drawdown,
- time period,
- markets traded,
- fees and slippage,
- and evidence that the trades actually occurred.
“The AI Generates Guaranteed Returns”
This should trigger immediate scrutiny.
Markets contain uncertainty.
Artificial intelligence does not remove that uncertainty.
The CFTC has specifically warned investors that AI cannot predict the future or sudden market changes and that fraudsters use promises of high or guaranteed returns to attract victims.
A recent example is particularly instructive.
In its September 25, 2026 complaint against Cash FX Group and others, the CFTC alleges participants were told expert traders, proprietary algorithms and AI could produce returns of up to 15% per week.
The agency alleges very little legitimate forex trading occurred and participants lost at least $406 million.
“The Backtest Proves the Strategy Works”
A backtest answers:
“How would this rule have behaved under these historical assumptions?”
It does not answer:
“What will this strategy earn next year?”
Important questions include:
- Was the strategy designed before or after looking at the data?
- Was there an out-of-sample period?
- Were fees included?
- Was slippage modeled?
- Were delisted assets included?
- Was liquidity realistic?
- Were thousands of parameter combinations tried before the winner was shown?
- Did the test include adverse regimes?
3Commas explicitly warns that historical backtest results are not a guarantee of future performance and notes that live trading can differ because of fees, liquidity, volatility and slippage.
“Demo Results Are Basically the Same as Live Trading”
Paper trading is extremely useful.
It is not live trading.
Coinrule's own current documentation provides a good example of transparent limitation disclosure.
Its demo exchange is designed to test strategy logic but does not currently account for live slippage in the same way.
Coinrule notes that demo performance can therefore be better than live performance, particularly on less-liquid assets.
Live markets introduce:
- slippage,
- partial fills,
- latency,
- spread changes,
- order-book depth,
- API delays,
- and exchange outages.
“Our Performance Is Audited or Verified”
The words audited and verified need objects.
Audited by whom?
What exactly was reviewed?
The company?
Its financial statements?
The trading algorithm?
A single exchange account?
A backtest?
A performance report?
Ask for:
- the auditor or verification provider,
- the exact legal entity reviewed,
- the reporting period,
- the methodology,
- whether trading records came directly from exchanges,
- whether deposits and withdrawals were distinguished from trading P&L,
- and whether the evidence is independently accessible.
A dashboard can be edited.
A screenshot can be manufactured.
A demo account can look profitable.
None of those independently establishes live trading performance.
“This Is Powered by Artificial Intelligence”
Ask the simplest possible question:
What exactly does the AI do?
Possible answers are very different.
It may:
- write strategy code,
- translate natural language,
- generate trading signals,
- choose parameters,
- rank predefined strategies,
- call external tools,
- or autonomously plan multiple steps.
Or the live bot may simply execute ordinary deterministic rules.
Cryptohopper offers an unusually clear example of why terminology matters.
Its established A.I. feature explicitly means:
Algorithmic Intelligence.
Its current documentation says that feature is not Artificial Intelligence.
It compares supplied strategies and selects among them.
That may still be useful.
But it is a different technology from an LLM or autonomous agent.
“Your Funds Always Stay Under Your Control”
This statement can mean several different things.
Ask where the assets actually sit.
Possible architectures include:
- assets remain on your centralized exchange,
- assets remain in your wallet,
- assets move into a smart contract,
- assets enter an agent wallet,
- assets are held by the platform,
- or assets are pooled with other users.
Then ask what the software can do despite not having custody.
A bot may be unable to withdraw your assets but still be able to:
- buy and sell,
- open perpetual positions,
- use leverage,
- move collateral inside an account,
- or create substantial trading losses.
“We Only Need Safe API Access”
There is no useful permission called:
“safe API.”
There are specific capabilities.
Check whether the credential can:
- read balances,
- read orders,
- place spot trades,
- trade derivatives,
- use leverage,
- transfer funds internally,
- withdraw externally,
- or access multiple subaccounts.
Also check:
- IP allowlisting,
- credential expiry,
- OAuth scope,
- revocation process,
- and whether separate credentials can be created for the bot.
Coinrule's current MCP documentation, for example, explicitly separates Read Only from Read + Write and allows connections to be revoked.
That is much more useful information than a general statement that the connection is secure.
“Trading Is Free” or “Fees Are Negligible”
A zero-dollar software subscription does not mean the trading stack costs zero.
Total friction may include:
- subscription fees,
- maker/taker fees,
- bid-ask spread,
- slippage,
- perpetual funding,
- borrow costs,
- AI model usage,
- API/data subscriptions,
- gas,
- bridging costs,
- and withdrawal fees.
High-turnover automation can amplify small friction.
The CFTC specifically advises investors evaluating AI trading systems to consider fees, spreads and subscription costs when judging returns.
“We Are Regulated, Licensed or SEC-Compliant”
Never verify a regulatory claim using the platform making the claim.
Use the regulator.
In the United States, depending on the activity involved, relevant checks can include:
- Investor.gov,
- SEC Investment Adviser Public Disclosure,
- FINRA BrokerCheck,
- and NFA BASIC for relevant derivatives and forex firms.
The CFTC recommends checking registration and disciplinary history before dealing with persons or firms that should be registered for relevant futures, commodity-pool, forex or derivatives activities.
The September 29, 2026 SEC complaints illustrate why this matters.
The SEC alleges that entities involved in the Cryptoaiml and TSAI schemes falsely claimed regulatory legitimacy, including through falsified SEC-related documents.
A regulator logo on a website is not evidence.
Neither is:
- an app-store listing,
- a certificate image,
- a Form D screenshot,
- a Telegram administrator,
- a famous person's photo,
- or an influencer endorsement.
“You Can Withdraw Your Money Anytime”
A trading dashboard showing $50,000 does not prove that $50,000 exists.
The real test is whether the balance can leave the system.
Investor.gov has warned that fraudulent trading platforms may display apparent profits and then demand:
- a tax,
- a processing fee,
- an additional deposit,
- or another payment
before supposedly releasing the money.
In the SEC's September 2026 Cryptoaiml complaint, the agency alleges investors attempting withdrawals were told their accounts were frozen until fraudulent advance fees were paid.
Before materially increasing exposure:
- test a small withdrawal,
- verify the destination,
- measure processing time,
- record the actual fee,
- and confirm that no unexpected additional payment is required to release your balance.
The DN Evidence-to-Claim Ratio
The 11 checks point toward a broader metric.
DN calls it the:
Evidence-to-Claim Ratio.
It asks:
How much independently verifiable evidence exists relative to the magnitude of the marketing claim?
Consider two companies.
Company A says:
“We provide rule-based automation. Here are the fees, API permissions, paper limitations and exact order logs.”
The promise is modest.
The evidence is detailed.
Company B says:
“Our autonomous proprietary AI generates 15% every week.”
Its evidence consists of screenshots and testimonials.
The second company has a radically lower Evidence-to-Claim Ratio.
The most trustworthy trading product may not be the one making the largest claim. It may be the one whose verifiable evidence most comfortably exceeds the ambition of its marketing.
What Counts as Stronger Evidence?
| Claim | Weak evidence | Stronger evidence |
|---|---|---|
| Performance | Screenshot | Reconstructable live trade history |
| Win rate | Marketing percentage | Trade count + payoff ratio + drawdown + period |
| Backtest | Single optimized chart | Methodology + costs + out-of-sample tests |
| AI | “AI-powered” label | Documented model role and workflow |
| Security | “Bank-grade security” | Permission architecture, custody model, revocation |
| Regulation | Logo or certificate | Direct regulator database record |
| Audit | “Verified results” | Named independent verifier and scope |
| Withdrawal | Dashboard balance | Successful real withdrawal |
DN AI Trading Claim Verifier
Score a trading platform before depositing or connecting a funded account.
AI Trading Claim Verifier
Evaluate eleven common marketing claims. “Verified” should mean you have independently checked the evidence, not merely read the company's own statement.
Do Not Deposit Yet
The $0 Verification Phase
A useful principle for AI trading experimentation is:
perform as much due diligence as possible before the platform receives any economic authority.
That means the first phase can often happen at:
$0.
Research:
- the company,
- the team,
- the legal entity,
- the documentation,
- the permissions,
- the pricing,
- the AI architecture,
- and the performance methodology
before funding anything.
Then use paper trading or demo functionality where available.
Two Current Examples of the Right Testing Order
Coinrule
Coinrule currently allows paper trading without connecting a live exchange.
Its MCP also separates Read Only and Read + Write access.
A user can therefore experiment in the order:
paper → read → review → write → live.
That does not make Coinrule risk-free.
It demonstrates useful permission sequencing.
Explore Coinrule3Commas
3Commas currently provides Demo Trading using virtual capital and real market data feeds.
That lets users understand bot behavior before a live exchange connection becomes the experiment.
Explore 3CommasThe Permission Before Capital Rule
There is a common mistake in trading automation:
users think about the amount they are depositing before thinking about the authority they are granting.
DN proposes reversing the order.
First ask:
- What can the software see?
- What can it change?
- What can it trade?
- Can it use leverage?
- Can it transfer assets?
- Can it withdraw?
- Can it create new approvals?
- Can I revoke it immediately?
Only then ask:
“How much money should I connect?”
In agentic finance, authorization may be a more important first variable than deposit size. A small account with unrestricted authority can be less controlled than a larger account behind strict deterministic limits.
The Screenshot Problem
A screenshot is evidence that a screenshot exists.
It does not prove:
- the account is real,
- the funds are real,
- the trades occurred,
- the performance belongs to the advertised system,
- the profits are withdrawable,
- or the image has not been edited.
That distinction matters even more now that generative AI can produce increasingly convincing:
- dashboards,
- testimonials,
- identity documents,
- videos,
- voices,
- and websites.
Investor.gov and the CFTC both warn that AI-generated media can be used to make fraudulent investment pitches appear legitimate.
The Three-Layer Verification Standard
For financially consequential claims, DN favors three levels of evidence.
| Layer | Question | Example |
|---|---|---|
| 1. Vendor evidence | What does the company say? | Documentation, terms, methodology |
| 2. External evidence | Can someone outside the company confirm it? | Regulator databases, audits, independent records |
| 3. User verification | Can I test the relevant claim myself? | Paper test, permission check, small withdrawal |
The strongest claims survive all three.
What Would Immediately Stop DN From Depositing?
Any one of these would justify stopping due diligence until resolved:
- guaranteed trading returns,
- pressure to deposit immediately,
- withdrawal access required without clear necessity,
- unknown legal entity,
- fake or unverifiable regulator certificate,
- profits that cannot be withdrawn without another payment,
- team members who cannot be independently verified,
- no explanation of what the AI actually does,
- screenshots presented as audited performance,
- refusal to explain fees,
- or a request to send funds directly to an individual's wallet.
The Zero-Dollar Burden of Proof
The company wants the deposit.
That means the company should bear the burden of proving the claims that justify the deposit.
DN calls this:
The Zero-Dollar Burden of Proof.
Before the first dollar moves, the customer should already understand:
who,
what,
where,
how,
how much,
and what can go wrong.
If the answers only become available after depositing, the information architecture itself is a warning.
Limitations
- Passing these checks does not prove a platform is safe or profitable. Due diligence reduces uncertainty; it does not eliminate risk.
- Regulatory requirements vary by product, activity and jurisdiction. Not every software tool is required to hold the same registration.
- Third-party audits have different scopes. An audited company is not the same thing as audited trading performance.
- Live performance can change. A genuine historical track record does not guarantee future results.
- Non-custodial systems still carry risk. Trading permissions, wallet approvals and smart contracts can create substantial exposure.
- AI terminology is inconsistent. Vendors may use AI to refer to technologies ranging from machine learning to simple optimization.
- Recent enforcement actions discussed here contain allegations. Where litigation is ongoing, allegations should not be treated as final judicial findings.
- DN does not certify platforms through this checklist. It is an educational due-diligence framework.
The Bottom Line
Do not start by asking:
“How much should I deposit?”
Start with:
“What exactly am I being asked to believe?”
Then separate every claim.
The win rate.
The return.
The backtest.
The AI.
The audit.
The custody model.
The API permissions.
The fees.
The regulatory status.
The withdrawal process.
And ask for evidence for each one.
A legitimate trading product should survive reasonable scrutiny.
A genuine AI product should be able to explain where the AI is.
A genuine trading record should be distinguishable from a demo.
A genuine regulatory registration should be visible through the regulator.
And a genuine account balance should eventually be withdrawable.
The correct amount to deposit while those questions remain unanswered is not:
$100.
Or $10.
Or $1.
It is:
$0.
DN built this article from current regulator guidance, recent enforcement actions and current first-party trading-platform documentation.
The framework evaluates eleven categories: Performance Claims, Guaranteed Returns, Backtesting, Demo-vs-Live, Performance Verification, AI Authenticity, Custody, Permissions, Costs, Regulatory Claims and Withdrawability.
The proprietary DN AI Trading Claim Verifier measures verification completeness rather than platform quality.
DN uses the concepts: Verification Burden Scaling, Evidence-to-Claim Ratio, Permission Before Capital and Zero-Dollar Burden of Proof.
Affiliate relationships contribute zero points and do not reduce the evidence standard applied to a platform.
Primary Sources & Evidence
- SEC, Charges Multiple Entities in Alleged Fraud Schemes Using WhatsApp, AI Signals and AI Trading Bots, September 29, 2026
- SEC Complaint, TSAI Pro Ltd. and TSAI Capital Foundation, September 29, 2026
- CFTC, Charges Cash FX Group and Others With Alleged $950 Million Fraud Scheme, September 25, 2026
- CFTC, AI Won't Turn Trading Bots into Money Machines
- Investor.gov, Artificial Intelligence and Investment Fraud Investor Alert
- CFTC, Check Registration and Backgrounds Before You Trade
- Investor.gov, Check Out Your Investment Professional
- Coinrule, Comparing Live Trading and Demo Exchange
- Coinrule, MCP Permissions and Security Explained, July 2026
- 3Commas, DCA Bot Backtesting Documentation
- 3Commas, Slippage in Algorithmic Trading, August 2026
- Cryptohopper, What Is Algorithm Intelligence A.I.?, September 2026
Frequently Asked Questions
Are 95% AI trading bot win rates real?
A high win-rate claim may be mathematically accurate while still hiding an unprofitable strategy. Verify total trade count, average win, average loss, drawdown, fees, timeframe and whether the results came from real trades rather than a backtest or demo.
Can an AI trading bot guarantee profits?
No trading AI can eliminate market uncertainty. The CFTC warns that claims of guaranteed or exceptionally high returns from AI trading systems are a major fraud warning sign.
How do I verify an AI trading bot's performance?
Look for reconstructable live trading evidence, clear treatment of deposits and withdrawals, full fees, trade-level records and an identifiable independent verifier where the company claims its performance is audited.
Are AI trading bot backtests reliable?
Backtests can be useful for testing strategy logic but depend heavily on historical data and assumptions. They can differ from live trading because of slippage, liquidity, fees, latency and changing market conditions.
Is paper trading the same as live trading?
No. Paper trading can test strategy logic without risking money, but simulations may not reproduce live liquidity, slippage, partial fills, latency or exchange conditions accurately.
How do I know whether a trading bot really uses AI?
Ask what the AI specifically consumes, produces and decides. A system may use AI for research, signal generation, natural-language translation, parameter selection or strategy development while the actual live trading remains deterministic.
Does non-custodial mean an AI trading bot cannot lose my money?
No. A non-custodial system may still have trading authority. A bot that cannot withdraw funds can still lose value through bad trades, leverage, liquidations or incorrect strategy execution.
Should a trading bot need withdrawal permission?
Ordinary exchange trading automation generally does not need withdrawal permission merely to place trades. Users should independently verify the exact permission requirements of the platform and exchange before connecting an account.
How can I verify whether an AI trading company is regulated?
Identify the exact legal entity and claimed regulator, then search the regulator's own database rather than relying on certificates, screenshots or links supplied by the promoter.
Should I test a withdrawal before depositing more?
Yes. Where funds are deposited with a platform, successfully completing a small withdrawal can verify an important part of the user experience before larger exposure is considered.
What is the DN Evidence-to-Claim Ratio?
The Evidence-to-Claim Ratio compares the amount and quality of independently verifiable evidence with the size and certainty of a product's marketing claims.
What is Verification Burden Scaling?
Verification Burden Scaling is the DN principle that stronger financial claims require proportionally stronger evidence. A claim of exceptional or guaranteed returns should require much more evidence than a simple claim about product functionality.
What is the Zero-Dollar Burden of Proof?
The Zero-Dollar Burden of Proof is the principle that a trading platform should provide enough evidence to understand its claims, costs, authority and limitations before the user transfers any money.
Freshness, Change Log & Corrections
| Date | Change |
|---|---|
| 8 October 2026 | Initial 2027 edition published with eleven pre-deposit AI trading verification checks. |
| 8 October 2026 | Added September 2026 SEC enforcement context involving alleged fake AI trading bots, fictitious profits and false SEC-compliance claims. |
| 8 October 2026 | Added September 2026 CFTC Cash FX enforcement context involving alleged AI and algorithmic trading claims and returns of up to 15% weekly. |
| 8 October 2026 | Verified current Coinrule demo-vs-live and MCP permission documentation. |
| 8 October 2026 | Verified current 3Commas backtesting and slippage disclosures. |
| 8 October 2026 | Verified Cryptohopper's current description of established A.I. as Algorithmic Intelligence rather than Artificial Intelligence. |
| 8 October 2026 | Added DN Verification Burden Scaling, Evidence-to-Claim Ratio, Permission Before Capital, Zero-Dollar Burden of Proof and AI Trading Claim Verifier. |
Last verified: 8 October 2026.
Correction policy: trading products, regulatory status and AI functionality change rapidly. DN will update this article when material claims, regulatory actions, verification methods or product architectures change.
Factual corrections can be submitted through the Decentralised News Contact page. Commercial relationships do not prevent corrections, negative findings or removal.
