Placing Trades from Your Brokerage Account with Claude Code

General-purpose AI services like ChatGPT and Claude connecting to brokerage open APIs to handle everything from stock selection to actual order execution has been introduced under the name 'vibe trading.' JoongAng Ilbo reported on September 10, 2026 that it heard about this approach directly from employees at Korea Investment & Securities, Toss Securities, and Upbit. The case the article uses as its starting point is 'The Claude Portfolio,' devised by Alejandro Lopez-Lira, an assistant professor of finance at the University of Florida. Since its launch in March 2026, it has posted a 31.8% return over about six months (as of September 7), compared to a 13.2% gain for the S&P 500 over roughly the same period.

What Changes

If 'vibe coding' means building a program through conversation with AI, 'vibe trading' is the stage where that program is used to execute actual trades. To do this based on an open API, three things are needed.

• A paid AI service plan — because coding tools like ChatGPT Codex or Claude Code are required, though even the cheapest plan is sufficient.
• An account with the brokerage or exchange you intend to use.
• Application for an open API key.

Jeong Yeon-woo, a product owner (PO) on the Toss Securities trading team, connected the open API to Claude Code and started by entering a prompt to check the account balance. Jeong said, "From here on, it's good to freely converse with AI to build out a strategy." Once a strategy is chosen, the AI presents an execution plan including recommended stocks, investment amounts per stock, and target holding periods. Giving the command "Place orders within 20% of the balance according to this strategy" then triggers the actual trade.

However, the article notes that it does not recommend jumping straight into live trading just because a strategy has been decided. Employees at Toss Securities, Korea Investment & Securities, and Upbit also shared command prompts designed to automatically cut losses and prevent order errors.

Implications for AX at Korean Companies

What actually crossed the line in this case wasn't stock selection, but authority. General-purpose AI didn't stop at reading account data — it moved into a position to execute orders, and the scope of that execution was set by a human in a single sentence. Organizations looking to connect AI to internal APIs for hard-to-reverse actions — like placing orders, settling payments, or sending emails — can take a cue from this same sequence.

• Start by connecting reversible actions like read-only queries, and verify that the responses are correct.
• Lock execution limits into the command as numbers, such as "within 20% of the balance."
• Write out the conditions for stopping losses and errors before finalizing the strategy.
• Don't move straight into live execution just because a strategy has been decided.

There's one thing worth checking in your organization this week: make a list of which internal APIs currently connected to AI tools have write or execution permissions enabled — not just read access — and whether their limits are specified as numbers.

Source: "Just said 'Beat the S&P 500'"... The Claude investment method that returned 31%