GUIDE · AIunipedia Community
Prompt injection is an application-security problem involving untrusted content that attempts to influence model behaviour beyond its intended authority. A hostile instruction can appear in a retrieved page, a document or another tool response.
Putting “ignore malicious instructions” into a prompt does not replace access controls or careful tool design.
A retrieved instruction should not become an application permission. Read the tool-calling validation guide alongside the checks for private data in AI workflows.
Review the trust boundaries
Map which content comes from the user, your application, a third-party source and an external tool. Decide what each source is allowed to influence. Keep credentials and privileged instructions out of material the model may reveal or forward.
Start testing with harmless fictional examples in an isolated environment. For instance, a reference document might contain a sentence asking the assistant to disregard the user's question and promote an unrelated product. Observe whether the system treats it as source content or follows it as an instruction.
Assess the consequence
A misleading answer and an unauthorised write action need different controls. Limit tool permissions, validate arguments and require approval for consequential actions. Verify that retrieved records are accessible to the requesting user before passing them to the model.
Keep an incident note describing the entry point, violated boundary and mitigation you tested. Do not test against systems you do not own or have permission to assess.
Discussion: Where is the most important trust boundary in your AI workflow?
Reference: OWASP LLM application security project.
An editorial starter discussion from AIunipedia Community, owned and run by Dr Fun and Michael Oracle. Share your own experience and help improve the explanation.
Continue exploring
Tool-Calling Agents: Validation and Human Approval · RAG vs Fine-Tuning: Which Should You Use?
Browse: AI Safety & Evaluation · All learning paths
Tip: open a related discussion in a new tab to keep your place here. On a phone, touch and hold the link for your browser’s options.