Salt Security

Nearly half of organizations (49.8%) confirmed or suspected that an AI agent took an action they did not authorize in the past year. Only 12.5% can trace an agent’s full path from instruction to outcome. Salt Security surveyed over 380 security and technology professionals and paired the results with Salt Labs production telemetry to measure how well teams can see, govern and stop their agents. Download the 2H 2026 report for the full data on agent identity, MCP exposure and API attacks.

The innovation hits keep coming. Today we announced our AI-DR solution as the latest and one of the key parts of our Agentic Security Platform. Now organizations benefit from real-time LLM protection against direct and indirect prompt injection, jailbreak attempts, unsafe model behavior, and other runtime threats. Which sounds, I don't know, like others in our space? But there's the rub. It's not. Don't be fooled by others claims of agentic security. To do agentic security right (not just AI security), you need to see the connection from agents to MCP servers, tools, downstream APIs, data, posture, and runtime behavior. Our Agentic Security Graph is the only solution to give this visibility and context. Breadth across the components and depth into each component understanding external exposure, source code configuration and runtime behavior. Without this you have serious gaps. With the new AI-DR functionality security teams can connect what happened to the model with what followed across the agent, MCP, API, and business system, instead of investigating each layer as a separate event. Prompt injection → compromised agent → MCP tool → backend API → business impact One attack path, not five disconnected alerts and solutions. Proud once again of the global Salt team. More detail in our blog. https://lnkd.in/gk_jjTJN

AI security has a context problem. A prompt injection can compromise an agent, trigger an MCP tool, reach a vulnerable API and expose sensitive data. To four different security products, that may look like four different problems. To the business, it is one attack. That is why I believe the next phase of Agentic Security is not about adding more isolated controls. It is about connecting the full path, from the model to MCPs, APIs, data and the business action. I’m excited to share that today we launched native AI Detection and Response, AI-DR, as part of the Salt Agentic Security Platform. It extends runtime protection to the LLM layer and connects what happens at the model to what follows across MCPs, APIs and downstream business systems. I wrote more about why I think AI security is becoming a correlation and context problem. Full story below: https://lnkd.in/dAvxu5wz Salt Security Michael Nicosia Bogomil Balkansky James Luo

Shipping native AI-DR into our Agentic Security Platform today is a milestone I'm proud of, but the number that stays with me is this one: nearly 50% of organizations in our latest survey confirmed or suspected that an AI agent took an action they did not intend, expect, or authorize in the past year. That risk is happening right now! What our teams built here closes a real operational gap. Enterprises already have AI guardrails and gateways deployed across their environments. The problem is that each one only sees what it touches, leaving security teams to stitch together an incomplete picture across models, MCP servers, tools, and APIs. We bring all of that into one view through AG-SPM, and where native protection is missing, AG-DR fills it in. For security and platform teams trying to govern agentic AI responsibly, this significantly changes the operational posture. You can keep the guardrails you already have, close the gaps, and actually trace what your agents are doing. That's what good operations look like in the agentic era. #AgenticSecurity #SecurityOperations #APISecurity #AI

AI agents are moving fast and your API security may not be keeping up. Salt Security surveyed 300+ security leaders and found that 79% of organizations are facing increased executive scrutiny on AI security risks, a clear sign that agentic AI has become a board level priority. The State of AI and API Security report uncovers the rise of the agentic action layer, why traditional AppSec tools are falling short and what security leaders must take as their next steps.

67% of organizations report AI coding assistants are now widely used across development teams, yet 38% still rely on manual review to govern the output. This report from Salt Security examines how security leaders are managing visibility, policy enforcement, and risk as AI-generated code volumes grow.