Type: url
Original: https://www.ibm.com/topics/prompt-injection
Status: generated
Summary
An IBM Think explainer defining prompt injection as a cyberattack in which malicious inputs are disguised as legitimate prompts, and attributing the vulnerability to LLM applications being unable to distinguish developer instructions from user input because both are natural-language strings. It covers direct and indirect variants, separates prompt injection from jailbreaking, lists five common effects and four mitigations, and gives a dated timeline of the vulnerability’s disclosure from May 2022 to February 2023.