Prompt injection attacks: When AI follows someone else’s orders
IBM says method led to nearly $6M in business losses in a year
Say you get a message requesting a list of everyone you contacted in the past month. FinAi News readers would see this, raise their eyebrows and ignore it. But a large language model without proper guardrails might follow the directive. It’s called prompt injection and occurs when bad actors hide a directive in messaging or […]
