Salt Security
salt.security ↗ · LinkedIn ↗ · Open on the map
AI CSPMAI Code SecurityRuntime & GuardrailsObservability & Governance
Capabilities
Agentic Security Platform (launched March 2026) built on Salt's API security product. The Agentic Security Graph maps how agents, LLM connections, MCP servers, tools and downstream APIs relate. Agentic Security Posture Management (AG-SPM) continuously discovers shadow agents, MCP servers and APIs and governs them against a Policy Hub of 100+ posture policies, many adapted from API posture rules. Agentic Detection and Response (AG-DR) flags misuse and abnormal behavior across that graph; AI-DR (September 2026) adds LLM-layer detection of direct and indirect prompt injection, jailbreaks and unsafe model behavior. Salt Code is an MCP server that feeds security policies (OWASP API, MCP and LLM Top 10) into AI coding assistants such as Cursor, Claude Code, Copilot, Kiro and Gemini CLI.
Sensors / integration
Agentless discovery first: cloud connectors (Salt Connect for AWS, Azure, GCP), external attack-surface scanning (Salt Surface), code repositories, and an integration with Claude Enterprise for MCP discovery. Runtime visibility comes from live API traffic collected through existing gateways, WAFs, load balancers and Kubernetes (Salt Collect; 40+ integrations including Cloudflare, Kong, Apigee and AWS). Existing AI guardrails and gateways are pulled into the same view rather than replaced. Salt Code runs as an MCP server inside the coding assistant.
Protections
Behavioral detection of API and agent abuse such as enumeration, logic attacks and sensitive-data exposure, tracing a request from the agent and prompt through MCP servers to the APIs and data it reaches. Blocking is pushed out through the customer's gateways and WAFs (Salt Protect, managed rules for AWS WAF) rather than an inline Salt proxy; AI-DR and AG-DR are described as detection, not inline prompt blocking. Posture findings drive drift detection and remediation of risky agent-to-API paths.
Differentiator
Comes at agent security from the API layer: it argues the main risk is what agents do through MCP servers and APIs, not the prompt, and reuses years of API traffic baselining to judge agent actions. Prompt-level runtime protection is newer and detection-oriented compared with AI-native guardrail vendors. Founded in Israel (as Secful) by Roey Eliyahu and Michael Nicosia, now headquartered in Palo Alto with R&D in Tel Aviv; last priced round was a $140M Series D led by CapitalG at a $1.4B valuation (2022).
Team & funding
- Founders
- Roey Eliyahu, Michael Nicosia
- Total raised
- $271M
- Investors
- CapitalG, Sequoia Capital, Advent International, Alkeon Capital, Tenaya Capital, DFJ Growth, Y Combinator, S Capital VC
- Previously
- Secful
Other AI CSPM companies
8Layers · AccuKnox · Act Security · Acuvity · AI Score · AIBound · Aiceberg · AIM Intelligence · Aim Security · Airia · Alert AI · Andromeda Security · See all →
Last updated · Get the “Listed on” badge