AI Governance Institute
← All news

Injection Attacks

Injection attacks are a class of exploits in which an attacker slips instructions into content that an AI system then reads and acts on, causing it to ignore its intended rules, leak data, or take unauthorized actions. For AI deployments, this includes prompt injection, where hostile text hidden in a document, email, or web page redirects a model that is connected to company tools. Because these attacks can turn an assistant with access to email, files, or payments into a channel for data theft, they are a core risk for general counsel, risk managers, and compliance teams to assess when approving AI tools.

1 item