New fraud trick: attackers hide money-transfer prompts in username fields of AI backends
kingslyj · x · 2026-11-18
A new type of fraud: attackers stuff long prompts into username/password form fields instructing AI-powered backends to transfer money to their bitcoin wallets, exploiting insecure backends that handle form submissions via LLMs. The poster quips that SQL injection "is so last century" — prompt injection is the new injection-class attack surface.
More from Safety
- Redditor argues AI isn't making software less secure — attackers using AI to find exploits is the real risk — WillGetBannedSoonn · 2026-11-18
- Open-sourced guardrail-middleware puts deterministic policy checkpoints between agents and their tools — MidnightHeavy3064 · 2026-11-18
- Alan Turing Institute warns UK increasingly exposed to AI disruption without sovereignty steps — turinginst · 2026-11-18
- Hinton on Fei-Fei Li's AI-risk 'hyperbole' critique: vested interests run AI companies — Formal-Grass-3173 · 2026-11-18
- Researcher and OpenAI Codex find Java hole in an hour, exposing Jira, Jenkins, Keycloak — LeviTurk · 2026-11-18
- Meet the Chinese start-up stress-testing top AI models for rogue behavior — pstAsiatech · 2026-11-18