Pillar Security protects the AI agents, models and workflows that enterprises are putting to work. As agents gain access to data, code and production systems, security teams need to know what is running and keep it within its business purpose. Pillar gives them one place to manage the risk of an “Agentic Workforce”.
It runs in the cloud, self-hosted or as a hybrid, inside the customer’s own VPC.
At A Glance
| Vendor | Pillar Security |
|---|---|
| Category | AI agent and AI application security |
| Identities covered | AI agents, MCP servers, coding agents, agent permissions and secrets in agent context |
| Deployment | Cloud, self-hosted or hybrid, in the customer’s VPC |
| Agentic AI | Discovery, posture, red teaming and runtime guardrails for AI agents and MCP |
| Customers | Similarweb, AvidXchange, Tavily, eleos |
| Certifications | SOC 2 Type II, ISO 27001 |
| Market classification | AI Model Security; also AI Safety & Guardrails |
What Pillar Security Does
Agents, models, MCP servers and tools are spreading through organisations faster than security teams can track them, much of it as shadow AI. Pillar Security follows the AI lifecycle in order. It inventories these assets and maps what each agent is told to do, which data it reads, which tools it may call and what it connects to. It then attacks the agents the way an adversary would, watches them at runtime to check they stay within their business purpose, and records the evidence auditors need.
The same lifecycle shapes SAIL (Secure AI Lifecycle), the company’s framework running from AI policy and discovery through posture, red teaming, runtime controls, sandboxing and governance, aligned with NIST, the EU AI Act, ISO and OWASP.
Products
AI Discovery And Posture
This module answers the first question most CISOs ask about AI: what is actually running. It builds an inventory of agents, models, MCP servers, skills, tools and coding agents, and highlights the ones that touch PII, code repositories, production databases or financial systems.
- Detects unapproved coding agents and personal automation tools connected to corporate data, including coding agents on developer laptops (Agentic Endpoint), before data leaves the machine
- Scores risk continuously, flagging over-broad permissions, missing human-in-the-loop controls and unrestricted write access to production
- Tracks drift from a known-good baseline and traces supply-chain dependencies
Red Teaming And Attack Surface Exposure (RedGraph)
RedGraph shows how an attacker could chain an agent’s tools, prompts and data into an attack path and how far the damage could spread. It simulates lateral movement through agent tools and skills, and runs multi-turn attacks, tool hijacking and indirect prompt injection against real agent configurations, with reproducible proof for every exploit it finds.
Runtime Guardrails
Runtime Guardrails stop attacks and data leaks while agents are working. Each agent gets adaptive guardrails that detect malicious intent and log activity. Taint analysis follows PII and secrets from source to destination and blocks unapproved outbound flows in real time, covering AI apps, agentic endpoints and AI gateways.
Governance And Compliance
For risk and compliance teams, this module enforces approved model lists, MCP allowlists and AI usage policies. It provides audit logs, compliance reports and policy lifecycle management for GDPR, CCPA and internal standards, and maps a single policy to OWASP, MITRE ATLAS, NIST and the EU AI Act.
How It Works
The platform deploys in the customer’s VPC with role-based access control, as cloud, self-hosted or hybrid. Discovery builds the inventory and maps permissions and connections; RedGraph tests against those real configurations; the runtime layer sits in the live request path, tracing sensitive data into outbound payloads, blocking flows and intercepting MCP error responses before they reach logs or clients; and governance turns policy into controls and reports.
Runtime checks can also run inside AI gateways. In Portkey, Pillar provides Scan Prompt and Scan Response checks for prompt injection, PII, secrets, toxic language and invisible characters. TrueFoundry AI Gateway calls Pillar’s guardrail endpoints before and after every LLM request.
Identities And Systems Covered
| AI agents | Agents acting outside their intended permission boundary, and agents that inherit their creator’s broad permissions instead of scoped access |
|---|---|
| Permissions | Wildcard permissions such as s3:*, missing human-in-the-loop gates and excessive agency |
| Secrets | .env files and AWS keys pulled into agent context, and credentials exposed in MCP error messages |
| MCP servers and tools | MCP inventory, allowlists and unapproved MCP connections |
| Platforms | AWS, Agentforce, GPT-4 and opencode |
Agentic AI Capabilities
Multi-Step And Multi-Agent Threats
Beyond single prompts, the platform watches chains of tool calls, so a file read followed by an outbound request can be flagged as possible exfiltration. In multi-agent orchestration it blocks poisoned instructions at the first handoff, and it inspects skill and tool definitions for dangerous shell execution patterns before they ever run.
Model Supply Chain
Models and packages are checked as they arrive. Unsafe model serialisation formats and unverified model publishers are caught and held for approval before deployment, and compromised packages are detected at download time when an agent pulls them in.
Company Background
| Founded | October 2023, in Tel Aviv |
|---|---|
| Founders | Dor Sarig and Ziv Karliner |
| Funding | $9M seed round (2025) led by Shield Capital, with Golden Ventures and Ground Up Ventures |
| Recognition | Included in the 2026 Gartner Cool Vendors in AI Software Security and the Frost and Sullivan Global Generative AI Security report |
Note: This overview is compiled from Pillar Security’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.