Salt Security Adds LLM Runtime Protection to Agentic Platform

Salt Security Adds LLM Runtime Protection to Agentic Platform

Salt Security has launched native AI Detection and Response (AI-DR) capabilities within its Agentic Security Platform, adding real-time protection for large language models against prompt injection, jailbreak attempts, and other runtime threats.

The new AI-DR layer sits within Salt's Agentic Detection and Response (AG-DR) product and connects LLM security to the company's existing API and agent monitoring. Security teams can now trace an attack from a malicious prompt through Model Context Protocol (MCP) servers to downstream API exploits within a single platform.

Connecting model attacks to business system risks

The integration addresses a specific vulnerability path: attackers manipulate an AI agent through prompt injection to expose connected tools and APIs, then exploit those systems directly or through the compromised agent. Salt's platform maps these connections using what the company calls the Agentic Security Graph, which links agents to their models, MCP servers, tools, and APIs.

According to Salt Security's survey data from the second half of 2026, nearly half of organizations (49.8%) confirmed or suspected unauthorized agent actions in the past year. Only 12.5% reported they could consistently trace an agent's full execution path from initial prompt through MCP servers to APIs.

The company provided a concrete example: an attacker could manipulate a billing agent to reveal information about a refund tool on an MCP server and its underlying API, then exploit that API to attempt unauthorized refunds. The platform would display this attack sequence from prompt injection through MCP server to API exploitation.

Unifying disparate AI security controls

Beyond native protection, the platform aggregates visibility across existing AI guardrails deployed in cloud platforms, endpoint solutions, SASE products, AI gateways, and managed AI services. Through the Agentic Security Posture Management (AG-SPM) component, teams can view guardrail configurations across environments and identify coverage gaps.

Where protection is absent—including for custom agents running in Kubernetes—Salt's native AI-DR fills the gap while preserving existing security investments.

The capabilities address prompt injection, which ranks as the top risk in the OWASP Top 10 for LLM Applications 2026. Salt detects both direct and indirect forms of prompt injection in real time.

Why it matters

As enterprises deploy AI agents with access to business-critical systems, the attack surface extends beyond the model itself. A successful prompt injection becomes a potential entry point to payment systems, customer databases, or internal tools. Security teams need unified visibility across this chain—from LLM to MCP server to API—to understand exposure and respond effectively. Salt's approach consolidates what has been fragmented across multiple security products and cloud services.

"An attack on an AI model can become an attack on the systems that run the business," said Roey Eliyahu, co-founder and CEO of Salt Security. "Our native AI-DR protects LLM interactions and connects what happens at the model to the tools and APIs downstream."

The AI-DR capabilities are available now within Salt AG-DR as part of the Salt Agentic Security Platform. Details were first reported by PR Newswire.