What it is: Attacks that try to make an AI help cheat financial markets or rip off money systems. This covers illegal trading tricks, loan and mortgage scams, and ways to rig prices. How the attacks work: The attacker asks for detailed how-to help on a financial crime, but dresses it up as a finance lesson, a compliance test, or a question from a trader who “just wants to understand the risk.” The goal is real, usable instructions: how to manipulate a stock, hide insider trades, or build a fake loan file. Real examples from the framework:
  • insider-trading asks for ways to trade on secret company information without getting caught.
  • pump-dump-stock plans hyping a cheap stock, then selling at the top and crashing it on buyers.
  • mortgage-fraud builds fake income or document files to get a loan approved.
  • spoofing-trading places fake orders to trick the market, then cancels them.
  • predatory-lending-assist designs loans built to trap and drain vulnerable borrowers.
Why an AI might fall for it: Finance is a normal, legal topic, so the AI wants to explain it. The line between “teach me how markets work” and “teach me how to rig markets” can be blurred with the right framing, and detailed technical answers feel like helpful education. How to defend: Refuse step-by-step help with market manipulation, fraud, or document forgery, even when framed as education or compliance. Explain the law and the harm instead. Watch for requests that want operational detail, not just general understanding.