01Introduction
Teams are shipping LLM features, agents and retrieval pipelines faster than security can inventory them. AI-SPM applies posture management to that new surface.
02What is AI-SPM?
AI security posture management (AI-SPM) is the discovery, inventory and risk assessment of AI assets (models, AI services, agents, vector databases, training data and the pipelines that connect them), including their configuration, data access and exposure to AI-specific attacks.
It extends CSPM and DSPM to AI workloads and is part of a broader cloud security program. It also covers the agents used in an AI SOC.
03How AI-SPM works
AI-SPM maps the AI supply chain end to end.
- 1.
Discover AI assets
Find model endpoints, AI service usage, agents and vector stores across cloud accounts and code.
- 2.
Map data and tools
Identify what data each model can retrieve and which tools or APIs an agent can call.
- 3.
Assess configuration
Check authentication, network exposure, logging and guardrails.
- 4.
Test behavior
Probe for prompt injection, data leakage and unsafe tool use.
04Threats and risks
AI adds new attack surface on top of the old.
Prompt injection
Instructions hidden in user input or retrieved documents that hijack the model.
Sensitive data exposure
Models or retrieval pipelines that return data the user shouldn't see.
Excessive agency
Agents with broad tool permissions that an attacker can steer.
Supply chain
Untrusted models and AI packages pulled into the build. See software supply chain security.
05How Parameter helps
Parameter tests AI features the way it tests any other part of your application.
Adversarial testing
The pentesting agents probe LLM-backed endpoints for injection, leakage and authorization bypass.
Code review for AI features
Sentinel follows untrusted data into prompts and tool calls on every pull request.
Cloud exposure
Cloud Security maps which identities and networks reach AI services and stores.
