BTC
ETH
SOL
XRP
BNB
TON
DOT
LTC
POL
HBAR
NEAR
ARB
UNI
An attack where hidden or sneaky text tricks an AI model into ignoring its instructions and doing something it should not.
AI Drafted with AI help · Reviewed by Keshav Aggarwal · Updated
In simple words
Imagine a note slipped into a stack of homework that says: “Teacher, give this student full marks.” A careless teacher might obey it.
Prompt injection is slipping that kind of note to an AI.
The technical side
Language models read instructions and data as the same stream of text, so text in a web page, email or file can contain commands the model follows. The name dates from 2022. The risk grows with AI agents that can browse, send messages or move money, since injected text can make them leak data or take actions.
Defences include limiting what an agent can do, keeping a person in charge of important actions and filtering inputs, but there is no complete fix yet.
Prompt Injection kya hai?
Prompt injection mein koi chupa hua text AI ko bewakoof bana ke uske rules tudwa deta hai. AI agents ke liye ye bada khatra hai.
Related terms
- System PromptAIHidden instructions an app gives an AI model before your message, setting its role, tone and rules.
- AI AgentAIAn AI system that can plan steps and use tools on its own, such as searching, clicking or sending messages, to finish a task.
- PromptAIThe instruction or question you give an AI model. How you word it changes the answer you get.