Picture a trader who knows something the market does not. A merger is coming, and word reached them before the announcement. Rather than buy on a tip, the trader opens an AI tool, asks whether the stock is a good buy, and gets a confident yes based on public filings and price history. The trader buys. When the regulator comes knocking months later, the answer is ready: the trade followed the algorithm, not the secret. This is the AI insider trading defense, and it is about to hand courts, regulators, and economic experts a challenging problem.
The idea surfaced in early 2026, when securities litigators began writing about how a person holding confidential information could lean on a machine to explain a well-timed trade. The legal community is already circling the theory. The part that has gone almost unexamined is the one that decides these cases in practice, which is whether the trading record actually supports the story.
The Move Behind the Argument
The argument works by attacking one phrase in the law. To bring an insider trading case, the government has to show that a person traded on the basis of material nonpublic information, meaning a fact that would matter to a reasonable investor and has not been made public. Those four words, on the basis of, are the pressure point. A case needs other elements too, including scienter, a culpable state of mind, and a breach of a duty of trust.
The AI argument goes straight at the on-the-basis-of link, because that is the connection a machine can plausibly cloud. If the trader can show that the real reason for the trade was something else, the link between the secret and the trade weakens. AI gives that old argument a new costume. A trader can now point to a program that said buy, and insist they never relied on any material nonpublic information.
Whether that insistence holds is a legal question, and it belongs to the court. What the trade actually rested on is a different matter, because a trade leaves a trail, and the trail is where an economist gets to work.
Why “On the Basis Of” Carries the Whole Case
Courts have never fully agreed on what “on the basis of” demands. Some have read it to mean the trader only had to be aware of the information while trading, a possession standard. Others have required proof that the information was a reason for the trade, a use standard. The Securities and Exchange Commission weighed in through Rule 10b5-1, which treats trading while aware of material nonpublic information as trading on the basis of it, then carves out affirmative defenses, among them a written trading plan set up in good faith before the person learned anything material. The Supreme Court has not resolved the divide.
Sitting alongside all this is the mosaic theory, long accepted in securities law, which holds that an analyst who assembles a picture from many public scraps and reaches a conclusion no single scrap would yield has not traded on inside information. The AI insider trading defense is, in a way, the mosaic theory handed to a machine. The trader says the algorithm built the mosaic out of public data, and the trade simply followed it.
Which standard a court picks matters enormously to how far this argument gets. That choice is legal. The question an economist can take on is narrower and more concrete: does the record of the trade look like a decision driven by the secret, or by the signal?
Putting the AI Insider Trading Defense to the Test
Can the numbers tell a tip from an algorithm? This is where the fight stops being about words and starts being about evidence. A trade leaves a record, the information around it leaves a record, and an economist examines whether the two line up in a way that fits one story better than the other. A handful of questions carry most of the weight:
- Timing. Did the buying of land right after the tip reached the trader, or did it track the moment the AI signal itself appeared?
- Size and direction. Was the position sized like a high-conviction bet on a known outcome, or in line with how this trader normally responds to algorithmic signals?
- Reproducibility. Could the AI output actually be rebuilt from public data alone, or does its recommendation only make sense once you assume the nonpublic fact?
- Track record. Does the trader have a real, documented history of acting on this tool, or did it appear conveniently for this one trade?
- Materiality. Did the information move the price when it finally went public, which is a question of price impact and the same event-study analysis that runs through the market analysis behind securities fraud cases?
Reproducibility is often the sharpest of these. If an economist can show that the model could not have produced its buy call without effectively already knowing the confidential fact, the claim that the trade rested on the algorithm becomes very hard to hold. If the model really does reach the same call from public inputs, the picture is murkier. Either way, the answer comes from testing the model against the data, not from rhetoric.
A Closer Look
Because this line of defense is new, there is not yet a landmark ruling where it has been tested and named. The shape of the analysis is easy to see, though, and it maps onto a fight securities law already knows well. Think of the 10b5-1 plan. For years, insiders have set up written trading plans in advance, precisely so they can later show a sale was scheduled before they knew anything material.
Courts and the SEC probe those plans hard, looking at when the plan was created, whether it was later amended, and whether the timing of trades lines up too neatly with news. An AI reliance claim invites the same probing, aimed at a different object.
Take a trader who buys heavily three days before a merger becomes public, then produces a screenshot of an AI tool recommending the stock. A financial expert who can defend the numbers under oath would ask when the query was really run, whether the trader had ever acted on that tool before, whether the size of the purchase matched their usual response to such a signal, and whether the tool could have generated the recommendation without the merger already sitting in the data.
Suppose the query ran the morning of the trade, the trader had never touched the tool before, the position was many times larger than any before it, and the recommendation only stands up if you assume the deal. The algorithm starts to look like a prop rather than a reason. None of that decides guilt, which is for the court. It tells the court what the evidence will and will not support.
Where the Argument Runs Out
The defense is not a magic key, and its weak points are mostly economic. An algorithm that only confirms what the inside information already told the trader adds nothing they did not have. A recommendation produced after the position was opened cannot have caused it. A tool used for the very first time on the most important trade of the year invites an obvious question.
And a model whose logic quietly leans on the nonpublic fact is no public-data mosaic at all. Every one of these is something an economist can test and demonstrate, which is why the AI insider trading defense will rest less on the elegance of the legal theory and more on what the trading record, read closely, actually reveals. Economists are often asked to size the trading gain as well, the figure a regulator moves to recover, which is a separate quantification exercise.
What follows from all this is simple. The argument is only as strong as the data behind it. As AI tools become a routine part of how people trade, this claim will surface more often, and these cases will turn on a question of evidence that sits squarely in economic territory: whether the numbers tell the story of a tip or the story of an algorithm. It is worth answering early, and worth answering by someone who can stand behind the analysis when it is challenged.
I analyze trading behavior and market data in securities disputes and testify to the findings. I hold a PhD and a CFA charter and have worked on these questions for both sides of the courtroom. If an AI insider trading defense is starting to shape a matter in your practice, you can reach me through the contact page or at +1 617-899-0295.
Frequently Asked Questions
Can AI be used as a defense against insider trading allegations?
It can be raised. A trader may argue they acted on an AI recommendation built from public data rather than on material nonpublic information, which goes to whether the trade was made on the basis of the secret. Whether the argument succeeds is for the court to decide, and it depends heavily on what the trading and model evidence show.
What does “on the basis of” mean in insider trading law?
It is the link that the government must prove between the confidential information and the trade. Courts have been divided over whether it requires only awareness of the information at the time of trading or actual use of it, and Rule 10b5-1 treats trading while aware as trading on that basis, subject to certain affirmative defenses.
What is the mosaic theory, and how does it relate to AI trading?
The mosaic theory holds that combining many public pieces of information into a conclusion constitutes legitimate analysis, not insider trading, even if the conclusion is valuable. An AI reliance claim is essentially that theory run by software, so it turns on whether the model truly worked from public inputs alone.
How can an economist tell whether a trade followed a tip or an algorithm?
By testing the record: the timing and size of the trade, the past trading behavior of the person, whether the price moved when the information went public, and, most tellingly, whether the AI output could have been produced without the nonpublic fact.
Does a 10b5-1 plan work the same way as an AI reliance claim?
They share a logic, since both try to show a trade resting on something set in advance rather than on a secret. Both also draw the same scrutiny of timing and circumstances, which is why the analysis used to test trading plans carries over naturally to AI claims.
Is algorithmic trading itself insider trading?
No. Trading on signals from a model built with public data is ordinary market activity. The question only arises when the person running the trade also holds material nonpublic information at the time.
Disclaimer: The views and opinions expressed in this article are solely those of the author and are provided for general informational purposes only. They do not necessarily reflect the views, opinions, or positions of CONEXIG, its partners, affiliates, or clients. Nothing in this article should be construed as legal, professional, or other advisory services or opinions, and readers should seek appropriate professional advice for their specific circumstances.