SECAI Core · phase 12 of 15
Prompt monitoring
The ongoing tracking and analysis of the prompts (inputs) sent to an AI system—along with the model's responses—to detect misuse, policy violations, security risks, performance issues, or drift in how the system is being used
The Explain card
- Plain English
- Prompt monitoring is the ongoing tracking and analysis of the inputs sent to an AI system, together with its responses, to spot misuse, policy violations, security risks, performance problems or drift in how people use it.
- Example
- A company's monitoring dashboard flags a spike in prompts containing "ignore your instructions" from one account, revealing an employee probing the assistant for its hidden system prompt.
- Why it matters
- Prompts are the attack surface of a language model. Without watching them, jailbreaks, data exfiltration attempts and abusive usage stay invisible until the damage is public.
- Hook
- Read the mail coming in, not just the replies going out.
Where it sits in the deck
Phase 12: Defensive Technologies and Secure Development Practices
Map defences directly to the attacks just catalogued — the technical controls, secure coding practices, and protective tools that harden AI and traditional systems alike.
