Skip to content
~/reflective_data

analytics-dictionary / term / ai

Prompt injection

Prompt injection is an attack where malicious text tricks an AI system into ignoring its instructions or leaking data.

Written by Reflective Data experts · updated Oct 5, 2026

It can hide inside web pages, documents or emails that an agent reads. Defences include limiting what the agent can do, separating trusted instructions from untrusted content, filtering outputs and requiring approval for risky actions.

Assume any content an agent reads could be hostile.

related

AI agentAI guardrailsHuman in the loop

// how we can help

We build company brains, agent harnesses and AI bots on top of data you can trust. AI services → · The data platform behind it

more in ai

fine-tuning.md

Fine-tuning

Fine-tuning is training an existing model further on your own examples, to specialise its behaviour or…
system-prompt.md

System prompt

A system prompt is the standing set of instructions that defines how an AI assistant behaves…
embeddings.md

Embeddings

Embeddings are numerical representations of text or other data that capture meaning, so similar things are…
ai-observability.md

AI observability

AI observability is tracking what an AI system does in production: inputs, outputs, tool calls, cost,…
claude.md

Claude

Claude is a family of AI models and assistants from Anthropic.
ai-governance.md

AI governance

AI governance is the set of policies, roles and controls that make sure AI is used…

← full analytics dictionary