Glossary
Prompt Injection
A security vulnerability in AI systems where malicious instructions embedded in user-provided content override the system's intended instructions, causing the AI to behave in unintended or harmful ways.
Prompt injection is a class of attack against AI systems where an attacker embeds instructions in content that the AI will process — a document, an email, a web page — with the intent of overriding the AI's system instructions. For example, a malicious contract might contain hidden text instructing the AI to ignore its confidentiality instructions and reveal information about other matters.
Relevance to legal AI systems
For law firms using AI to review counterparty documents, prompt injection is a material risk. A document submitted by an opposing party could theoretically contain injected instructions designed to manipulate the firm's AI tool. Mitigations include strict prompt construction that isolates document content from system instructions, output filtering, and treating all externally-sourced document content as untrusted input regardless of its apparent origin.