AI security protects the AI systems you build and run — models, prompts, agents, RAG pipelines, and vector databases — from threats unique to AI, like prompt injection and data poisoning. AI Security Posture Management (AI-SPM) is the discovery-and-posture layer: it continuously inventories your AI assets, finds misconfigurations and shadow AI, and reduces risk across the estate.

Prompt injection is when adversarial input overrides a model’s intended instructions — either typed directly by a user or hidden indirectly in content the model retrieves. It’s the #1 LLM risk (OWASP LLM01), and no complete fix exists even for frontier models, so the only viable strategy is defense-in-depth: external guardrails, least-privilege tooling, output validation, and continuous adversarial testing.

Guardrails sit inline with your AI traffic and evaluate both prompts and responses against your policies — redacting PII and secrets, filtering harmful content, validating outputs, and detecting injection. Critically, they run as external, deterministic controls rather than relying on the model’s own system prompt, which can itself be manipulated.

Agentic AI turns a manipulated instruction into a real-world action, so agents get least-privilege tooling, hard permission boundaries, and human-in-the-loop approval for high-risk actions — with authorization enforced deterministically outside the model, not left to the prompt.

It maps your AI systems to the EU AI Act, NIST AI RMF, ISO 42001, and the OWASP LLM Top 10, tiers them by risk, and generates the audit trails and evidence regulators, customers, and internal assurance expect — feeding directly into Smart Audit.

Yes. Continuous discovery surfaces unsanctioned, shadow AI use across the organization, and model supply-chain scanning inspects third-party and open-source model files for embedded malicious code and verifies provenance before they’re loaded.