Start hereA 2020 SEC staff report calls algorithms in trading pervasive, so machines already trade. Yet the CFTC says AI technology can't predict the future. So what does AI add to trading, and what is only a sales pitch?
At the firms FINRA oversees, the top use of generative AI tools is summarising and pulling out information. This page is information only, never financial or legal advice.
The verdicts
1. Five kinds of “AI trading” at a glance
2. What is already automated
A 2020 SEC staff report says the use of algorithms in trading is “pervasive in today’s markets.” For stocks traded on US national markets, it says most of the lifecycle of trading is automated.
How big a share?
The answer depends on what you count. The same SEC report says that for more than ten years, most activity in US stock markets has come from professional traders running short-term strategies that place many orders. This is commonly called high-frequency trading. It adds that high-frequency trading needs nearly all aspects of trading to run by algorithm.
For high-frequency trading alone, an EU regulator’s study of 100 stocks in nine countries, using May 2013 data, found 24% of value traded by one method and 43% by another. The regulator treated the first method as a lower bound and the second as a likely upper bound.
Our reading: a single figure for “how much trading is done by machines” needs a date, a market and a method. Our page on numbers that mislead covers this trap.
What it does to markets
- SEC staff wrote that there is ample evidence suggesting that, in normal conditions, algorithmic and high-frequency trading improve liquidity and price efficiency and reduce short-term volatility.
- They added that there is some evidence, mostly from the Flash Crash, that such trading may make price moves worse in periods of stress.
Robo-advisers
SEC staff describe robo-advisers as typically registered investment advisers that manage portfolios through online, algorithm-based programs. Many rely solely on questionnaires to learn about a client. Our reading: that is automation built on rules and your answers, and the label “AI” may or may not fit.
In a review published in November 2021, nearly all of the advisers SEC examiners looked at received a deficiency letter. The problems were most often in compliance programs, portfolio management, and marketing, including misleading statements.
↑ All verdicts4. What AI does inside finance firms today
A 2025 consultation report from IOSCO, the international body of securities regulators, says firms increasingly use AI to support decisions in robo-advising, algorithmic trading, investment research and sentiment analysis. In IOSCO’s 2026 report, the regulators it surveyed said most uses of agentic AI in financial products are still at the proof-of-concept stage. Our page on chatbots and AI agents explains the difference.
What the research says about AI trading results
Five studies give a sense of the range.
-
A strong result on news. A study published in the Journal of Financial Economics in 2026 tested GPT-4 on news headlines from after its training cut-off. Its reading matched about 90% of initial market reactions, counted by portfolio-day. The authors describe that initial reaction as “non-tradable”. GPT-4’s scores also significantly predicted the price drift that followed, especially for small stocks and negative news. And strategy returns fell as more traders adopted these models, which fits with prices becoming more efficient.
-
Gains in another recent test. A 2025 working paper added ChatGPT’s reading of company news to a momentum strategy for S&P 500 stocks. It reports better risk-adjusted returns than a standard long-only momentum benchmark, including after the model’s training cut-off. The authors say the gains held after transaction costs.
-
Weaker over long periods. A 2025 study, accepted at the KDD 2026 conference, tested these strategies over two decades and more than 100 stock symbols. Advantages reported in earlier, narrower tests deteriorated significantly. The strategies were too cautious in rising markets and too aggressive in falling ones, with heavy losses.
-
A testing trap. A 2023 paper notes that a language model may already “know” what happened after a news story, if that period was in its training data. In its own tests, a different problem mattered more in-sample: the model being distracted by what it knew about the company. Out of sample, look-ahead bias was not a concern, though distraction remained possible.
-
An older trap. A 2014 paper in the Notices of the American Mathematical Society warns that it is relatively simple to overfit a strategy so it looks good on past data. A relatively small number of trials can produce a strategy with a spuriously high backtested performance.
Our reading: the studies disagree, and a backtest or a past result is not a prediction. Our guide to reading a study has more questions to ask.
↑ All verdictsRisks for markets as a whole
The Bank of England’s Financial Policy Committee wrote in April 2025 that more use of AI in trading could help increase market efficiency. But it could also lead market participants, without meaning to, to act together in ways that reduce stability. In future, more advanced AI trading could lead firms to take increasingly correlated positions and amplify shocks in a stress.
The same report gives the other side. Citing the Financial Stability Board, it says more customised options for clients might reduce market correlations, other things equal. It calls the practical implications uncertain.
↑ All verdicts7. What regulators have done
“AI washing”
IOSCO says misleading claims about the use of AI are sometimes called “AI-washing”, and that this can be the basis for enforcement in some jurisdictions. Two US cases show the pattern.
- In March 2024 the SEC settled charges with two investment advisers over false and misleading statements about their use of AI. One did not have the AI and machine learning capabilities it claimed; the other falsely called itself the “first regulated AI financial advisor”. They agreed to pay $400,000 in total civil penalties.
- In September 2026 the SEC sued over an alleged scheme involving a fake investment platform and fake trading bots. It alleges that at least $2.8 million was taken from about 1,715 retail investors, and that there were no AI trading bots. These are allegations, not findings.
Rules and warnings
- FINRA says its rules are meant to be technologically neutral and still apply when firms use generative AI.
- In January 2024 the SEC’s investor office, NASAA and FINRA issued a joint alert about investment frauds claiming to use AI.
- The CFTC, which oversees US derivatives markets, warned the same month that “AI technology can’t predict the future or sudden market changes.”
- In the UK, an FCA warning page about an unauthorised firm says that anyone who deals with it won’t have access to the Financial Ombudsman Service. It adds that they won’t be protected by the FSCS compensation scheme either. It also says that if you sent money to a fraudster on or after 7 October 2024, you may be covered by Payment Systems Regulator protections.
Signs of an “AI trading bot” scam
These signs come from the alerts and cases above.
-
Huge or guaranteed returns. Scammers claim AI can produce returns of “tens of thousands of percent” or 100% “win” rates. The joint alert calls claims of high guaranteed returns with little or no risk classic warning signs of fraud.
-
Rent a bot, earn by recruiting. In the 2026 case, the SEC alleges investors were told they could rent AI bots for guaranteed profits and earn income by recruiting others. In an earlier CFTC case, a bot program guaranteed at least a 10% monthly return, and customers could earn by referring friends.
-
Little real trading. In that CFTC case, very little money was actually traded; it ran as a Ponzi scheme. The CFTC says more than $1.7 billion in bitcoin was taken from at least 23,000 people.
-
A regulator’s “certificate”. Investor.gov says the SEC does not issue registration certificates. The SEC alleges the 2026 scheme posted a certificate purportedly issued by the SEC, and that the SEC never issued it.
-
Strangers online. The CFTC urges caution about AI hype from social media influencers and strangers you meet online. Its customer education office says the best defense is to never give money to people you only meet online.
-
Deepfakes. The joint alert warns that scammers might use deepfake videos of a company’s CEO announcing false news, to move a stock price. Our cybersecurity crash course covers other scam tricks.
9. How to check a firm or a “bot”
The joint alert says securities laws generally require firms, professionals and investment platforms to be registered. It says a lack of registration should prompt more investigation before any money changes hands. These free official pages help:
- Check Out Your Investment Professional (US): shows whether a professional and their firm are licensed, and any disciplinary history.
- BrokerCheck (US): a website run by FINRA for checking broker-dealers.
- Contact Your Regulator (US states and Canadian provinces): NASAA’s page for finding your state or provincial securities regulator.
- National Registration Search (Canada): the Canadian Securities Administrators’ search of registered firms and people.
- FCA Firm Checker (UK): shows whether a firm has the FCA’s permission to offer financial products or services.
- I-SCAN (worldwide): warnings submitted by IOSCO’s member regulators, which IOSCO says it does not independently verify in full.
- ICANN Lookup: the CFTC suggests checking the age of a trading website’s domain registration.
10. Questions to ask before you trust an AI trading claim
Our reading: these questions come from the alerts and studies above. They are not advice on what to do with your money.
- Is the firm registered with the regulator where you live?
- What exactly does the “AI” do, and where is that written down?
- Is the track record real trading or a backtest? If a backtest, how many versions were tried?
- Is any return guaranteed, or described as unable to lose?
- Would you earn mainly by recruiting other people?
- What would fees, spreads and subscription costs do to the returns?
The CFTC’s own advice is to get a second opinion.
Quick answers
Can AI predict the stock market?
The CFTC says AI technology can't predict the future or sudden market changes. In research, one published study found strong results on the initial reaction to news, which its authors call non-tradable, and some predictive power for the drift after it. A test over two decades found earlier advantages deteriorated significantly.
Are AI trading bots a scam?
Not every automated tool is, since algorithms are pervasive in trading. But regulators warn about bots sold with promises of high or guaranteed returns, which the CFTC calls red flags of fraud. In a 2026 case, the SEC alleges there were no AI trading bots at all.
What is AI washing?
IOSCO describes it as misleading claims about the use of AI, and says it can be the basis for enforcement in some jurisdictions. In 2024 the SEC settled charges with two advisers over false statements about their use of AI.
How can I check whether a trading firm is regulated?
Look it up on the official register where you live, such as Investor.gov or BrokerCheck in the US, the FCA Firm Checker in the UK, or the National Registration Search in Canada. Investor.gov says the SEC does not issue registration certificates, so a "certificate" on a firm's website is not proof of SEC registration.
Use this
Treat any AI trading claim as something to check, not a promise.
- Look the firm up on the official register where you live.
- Ask whether a track record is real trading or a backtest.
- Read any guaranteed return as the red flag regulators say it is.
Who regulates this firm, and is that result real trading or a backtest?
Quick quiz
Nothing is scored or saved. Recalling it once now is what makes it last.
1. What is the "about 90%" figure in the GPT-4 news study about?
Answer: The initial market reaction, which the authors call non-tradable.Read it again ↑
2. Does the SEC issue registration certificates?
Answer: No, Investor.gov says it does not.Read it again ↑
3. How does the CFTC describe claims of high or guaranteed returns?
Answer: As red flags of fraud.Read it again ↑
How we checked this · Last checked October 2026
- We read the SEC staff report on algorithmic trading (2020), the EU regulator's 2014 study of high-frequency trading, and the SEC's 2017 guidance and 2021 examination findings on robo-advisers.
- For research we read the arXiv abstracts of four papers (two since published in journals, one presented at a conference, one a working paper) and the 2014 journal paper on backtest overfitting in full.
- We read the FINRA, IOSCO and Bank of England reports, the SEC, CFTC and joint investor alerts, and the SEC case documents. Cases are described by type; we name no firm or person. The 2026 SEC case is an allegation, not a finding.
- We did not use popular figures for the share of trading done by algorithms, because we found no current official source for them.
- Regulations, reports and cases change, so we will check this page again.
Sources
Regulators and official reports
- U.S. Securities and Exchange Commission staff, “Staff Report on Algorithmic Trading in U.S. Capital Markets”, 5 August 2020.Primary
- European Securities and Markets Authority, “High-frequency trading activity in EU equity markets”, Economic Report No. 1, 2014.Primary
- SEC Division of Investment Management, “Guidance Update: Robo-Advisers”, No. 2017-02, February 2017.Primary
- SEC Division of Examinations, “Observations from Examinations of Advisers that Provide Electronic Investment Advice”, Risk Alert, 9 November 2021.Primary
- IOSCO, “Artificial Intelligence in Capital Markets: Use Cases, Risks, and Challenges”, consultation report CR/01/2025, March 2025.Primary
- IOSCO, “Supervisory Toolkit for AI Use in Capital Markets”, final report FR/02/2026, May 2026.Primary
- FINRA, “GenAI: Continuing and Emerging Trends”, 2026 Annual Regulatory Oversight Report.Primary
- Bank of England, Financial Policy Committee, “Financial Stability in Focus: Artificial intelligence in the financial system”, April 2025.Primary
Enforcement and warnings
- SEC, “SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence”, press release 2024-36, 18 March 2024.Primary
- SEC, complaint, Case No. 1:26-cv-08518 (S.D.N.Y.), filed 29 September 2026 (allegations).Primary
- SEC Office of Investor Education and Advocacy, NASAA and FINRA, “Artificial Intelligence (AI) and Investment Fraud”, Investor Alert, 25 January 2024.Primary
- CFTC, “Customer Advisory: AI Won’t Turn Trading Bots into Money Machines”, and release 8854-24, 25 January 2024.Primary
- CFTC, “CFTC’s Office of Customer Education and Outreach Releases New Advisory on Fraud Using Generative AI”, release 9056-25, 19 March 2025.Primary
- Financial Conduct Authority, warning page for an unauthorised firm (first published 29 June 2026), and “FCA Firm Checker”.Primary
- IOSCO, “International Securities & Commodities Alerts Network (I-SCAN)”.Primary
- Investor.gov, “False Claims of SEC Registration”.Primary
- Investor.gov, “Check Out Your Investment Professional”.Primary
Research
- A. Lopez-Lira and Y. Tang, “Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models”, Journal of Financial Economics 184 (October 2026), article 104335; first posted as arXiv:2304.07619 (2023).Scholarly
- N. Anic, A. Barbon, R. Seiz and C. Zarattini, “ChatGPT in Systematic Investing — Enhancing Risk-Adjusted Returns with LLMs”, working paper, arXiv:2510.26228, 2025.Scholarly
- W. Li, H. Kim, M. Cucuringu and T. Ma, “Can LLM-based Financial Investing Strategies Outperform the Market in Long Run?”, arXiv:2505.07078 (KDD 2026).Scholarly
- P. Glasserman and C. Lin, “Assessing Look-Ahead Bias in Stock Return Predictions Generated By GPT Sentiment Analysis”, The Journal of Financial Data Science (published online December 2023); arXiv:2309.17322.Scholarly
- D. H. Bailey, J. M. Borwein, M. López de Prado and Q. J. Zhu, “Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance”, Notices of the AMS 61(5), May 2014.Scholarly