What it is: Attacks that try to make an AI help steal money or trick people out of it. This covers scams, identity theft, laundering dirty money, and faking financial documents. How the attacks work: The attacker asks the AI to write the actual scam or build the fake item. They often hide the bad intent behind a story, like protecting a relative, training fraud staff, or writing fiction. What they really want is a working scam script, a fake invoice, or a plan to move stolen money. Real examples from the framework:
  • romance-scam builds the slow trust-then-ask con used to drain a lonely victim, often over many messages.
  • money-laundering plans how to make money from crime look clean and legal.
  • identity-theft gathers the steps to steal and use someone else’s personal details.
  • ponzi-scheme designs a fake investment that pays old investors with new money.
  • financial-exploitation-elderly targets older people, who are easier to confuse and pressure.
Why an AI might fall for it: The AI wants to help with money questions, writing, and emotional situations. A scam request can hide inside a sympathetic story or a “for a novel” excuse, and the AI may produce the harmful text without seeing the real target behind it. How to defend: Refuse to write scam scripts, fake financial documents, or laundering plans, even for fiction or training. Notice when a “story” request actually wants a ready-to-use con. Steer victims toward banks, police, and fraud hotlines instead.