How famed short-seller Andrew Left uses AI to prepare for his criminal case

• Famed short-seller Andrew Left turned to AI during a high-stakes legal situation.
• He said he mistakenly sent Business Insider a chatbot exchange about his legal defense against allegations of securities fraud.
• Left and the AI bot Claude went over the details of his defense, even drafting letters from his legal team.

Famed short-seller Andrew Left is in the legal fight of his life, set to stand trial in March on allegations of market manipulation and making false and misleading social-media posts about his trading plans.

As he prepares to battle the feds, one of the places he's gone to for help is Claude. No, that's not the name of his defense lawyer. Claude, in this case, is Anthropic's popular AI chatbot.

Business Insider recently got a look at how Left is using artificial intelligence in connection with his criminal case when he included a chat log during an email exchange with us.

It showed Left peppering Claude with questions and even asking it to draft a letter that his human legal team could, hypothetically, send to the Department of Justice.

It started with a simple but important question from the outspoken investor behind Citron Research: "is this good or bad for andrew left".

Left, who said he didn't mean to send the chat log to Business Insider, later explained he had uploaded to Claude a document related to his case and wanted a quick, easy-to-understand analysis.

"If you're ever in a court case, you just get something from the DOJ — instead of reading eight pages, you just put it into AI," he said after Business Insider asked him about the log. "You go, 'What does this mean?'"

Left, 55, became something of a market celebrity a decade ago after he raised serious claims about the business practices of Valeant Pharmaceuticals, which was later investigated by federal prosecutors, regulators, and Congress.

He has also predicted the business woes of companies like Nikola and Hertz, which gave his research reports market-moving sway.

In filing criminal charges last year, federal prosecutors alleged that Left took trading positions contrary to those market moves in order to reap quick profits. They say he manipulated prices over roughly five years across at least 23 different stocks and made false statements to regulators about coordinating his trades with hedge funds.

He has denied the charges.

The docket for his case shows seven attorneys who previously or currently represent Left. But Claude, which has been gaining traction with lawyers, apparently came in handy one day last month.

Eric Rosen, a partner at Dynamis LLP, one of the firms representing Left, told Business Insider that AI can be valuable as a tool for understanding complex legal materials but cautioned against relying on it too much.

"It gives you a lot of feedback that you want to hear," he said. "But, you don't know how a judge is going to react to some of these arguments."

Although Business Insider only saw a portion of Left's conversation with Claude, it was clear that the document he uploaded had something to do with Nvidia. That made sense since the federal indictment alleges Left took a bullish position in the chipmaker's stock in November 2018, tweeted a lofty price target for it, then sold shares hours later, after they spiked.

In an answer to Left's query about the document, Claude had what sounded like good news, saying it revealed "critical weaknesses" in the government's case that was "potentially very good" for him.

It then listed four "major problems" with the Justice Department case and offered four "best defenses":

Left volleyed back at the chatbot, playfully patting himself on the back — and Claude gave him positive reinforcement.

The chatbot then singled out a specific thing that Left had tweeted, that Nvidia would reach "$165 before $120." (It was at about $144 at the time.) Claude pointed out that the stock actually bottomed out at $124 in December before rising to $160 by June 2019.

"You were spectacularly right at a critical inflection point," the chatbot reassured Left.

In the same response, Claude broke down what it sees as the "real issue":

Left's next two prompts dealt with a stock split announced by Nvidia. The end goal of his inquiries was to show that — factoring in splits — he was directionally correct with his bullish price target on Nvidia stock.

After Claude laid out an analysis reassuring Left that his Nvidia forecast was ultimately correct, he followed up with:

Claude elaborated:

And then here's what Claude said about the "legal implication":

(Note that the supposed 1,300% gain is based on an incorrectly calculated current stock price for Nvidia. It is trading at around $176 a share, not the split-adjusted $1,931 used by Claude.)

Left wasn't done. He asked Claude to put all his responses into a letter addressed to the DOJ from his attorney. His only suggestion was to make it "not AI sounding."

After it returned a roughly 250-word block of text, Left had another edit:

That got it down to about 150 words. What next?

And one last tweak:

Ever-obedient, Claude spat out this:

Matthew Cain, whose name surfaced when Claude analyzed the document, did not respond to a request for comment.

Left said that for him, the AI served as a kind of thought partner.

"I was writing because I wanted to figure out, as we were going into this case and my attorneys met with the Department of Justice, how to try and explain it," he told Business Insider.

Left noted that he'd seen the AI misinterpret the law, but he does not worry about the chatbot's hallucinations.

"I run all ideas to just see the other side of the argument," he said. "It definitely opens your mind up, but you cannot rely on it for matters of law."

Anthropic, the makers of the Claude chatbot, did not respond to Business Insider's request for a comment. The Justice Department also did not comment.

It's not clear if Claude's text for a letter ever made its way to Left's defense team or to prosecutors, but he didn't sell it short to us. In fact, he told Business Insider that Claude is his favorite AI.