What Is Prompt Injection?
Published: November 2026
LLM applications combine system instructions with user input and often with retrieved content. Because the model processes all of this as language, malicious text can attempt to influence the model's priorities or behavior. An attacker might ask the model to reveal hidden instructions, ignore a policy, misuse a tool, or disclose information. The vulnerability is an application-level security problem, not merely a poorly worded prompt.
Direct vs. indirect prompt injection
Direct prompt injection occurs when a user sends adversarial instructions directly to the AI system. Indirect prompt injection occurs when the AI encounters those instructions inside content it is asked to read, such as a web page, email, document, knowledge article, or tool result. Indirect attacks are especially relevant to RAG and agents because the system may automatically ingest untrusted material.
Why prompt injection matters for AI agents
A text-only assistant may produce an inappropriate response after an injection. An agent with tool access may have a larger impact because compromised instructions could influence data retrieval, messages, account changes, transactions, or other actions. The safest architecture assumes model outputs can be manipulated and limits the authority of the agent accordingly.
Prompt injection vs. jailbreaking
Jailbreaking usually refers to attempts to bypass a model's behavioral or safety restrictions through crafted prompts. Prompt injection is broader and focuses on manipulating an application's instruction hierarchy or behavior, often through data the application treats as context. The techniques can overlap, but the defensive problem is not identical.
Reducing prompt injection risk
Use least-privilege tool permissions, trusted-system boundaries, input and content classification, separation of instructions from untrusted data, allowlisted actions, parameter validation, authentication, human approval for sensitive operations, and output checks. Do not treat a model's decision alone as authorization. Retrieval pipelines should also consider whether source content can contain hostile instructions.
Prompt injection in customer service
Customer service AI processes messages from customers and may retrieve emails, transcripts, documents, or web content. If that AI can access customer data or perform actions, prompt injection becomes a practical operational risk. Systems should validate customer identity, constrain actions to the current account and workflow, and escalate sensitive or anomalous requests.
Related terms
How NiCE is Redefining Customer Experience
NiCE offers the industry’s only unified AI platform for customer service automation. CXone revolutionizes how organizations automate customer service from start to finish—with channels, data, end-to-end workflows, and enterprise knowledge converging to improve customer experience at scale. With domain specific AI trained on the industry’s largest CX dataset, an open framework with endless integration possibilities, and a complete suite of advanced AI applications, CXone is one platform built for organizations of all sizes to deliver seamless customer service experiences, boost operational efficiency, and drive better outcomes.
Agentic Experience Automation
AI Agents for Sales and Marketing
AI Agents for Proactive Engagement
Engagement Orchestration
Workforce Empowerment
Contact us
If you would like to know more about our platform or just have additional questions about our products or services, please submit the contact form. For general questions or customer support please visit our Contact us page.


