Salt Security has revealed new native AI Detection and Response (AI-DR) capabilities within the Salt Agentic Security Platform.
According to the company, the new capabilities provide real-time LLM protection against direct and indirect prompt injection, jailbreak attempts, unsafe model behavior and other runtime threats.
Prompt injection is the top ranked risk in the OWASP Top 10 for LLM Applications 2026, taking on two forms, direct and indirect, both of which the new capabilities detect in real time.
Salt said that this also brings the AI guardrails and gateways enterprises already use into one security view, helping teams apply consistent protection across models, agents, MCP servers, tools and downstream APIs.
AI agents connect models to tools and business systems.
An attack that starts with a malicious prompt can expose those connections and lead to unauthorized access or actions downstream.
Salt added that it is not hypothetical: in Salt Security’s 2H 2026 State of Agentic AI and API Security survey, nearly half of organizations (49.8%) confirmed or suspected that an AI agent took an action they did not intend, expect or authorize in the past year and only 12.5% can consistently trace an agent’s full path from the initial prompt through the MCP servers and APIs it reaches.
Security teams need to see how an attack moves across these layers to understand what is at risk.Â
AI-DR is built into Salt Agentic Detection and Response (AG-DR), extending its runtime protection to the LLM layer.
The Salt Agentic Security Graph connects agents to their models, Model Context Protocol (MCP) servers, tools and downstream APIs.
With native AI-DR, teams can connect an attack against the model to subsequent MCP and API attacks in the same platform.
AG-DR and Agentic Security Posture Management (AG-SPM) form the Salt Agentic Security Platform.
The company added that AG-SPM helps teams understand their agents and security gaps, while AG-DR detects and connects attacks across those agents and the systems they use.
For example, an attacker could trick a billing agent into revealing information about a refund tool on an MCP server and the API behind it.
The attacker could then exploit that API – directly or through the agent to attempt unauthorized refunds.
What began as a malicious prompt has become an attack on a payment system.
Salt shows these steps together, from the prompt injection through the MCP server to the API attack.
Security teams can see how the attack started, what it targeted and which business systems are at risk, even when the attacker moves beyond the agent to attack an API directly.