Pillar Security

Pillar Security is an Israeli AI security startup selling one platform that spans the AI application lifecycle. It starts with discovery: automatically mapping an organization's AI assets into a live AI-BOM, covering models, datasets, prompts, notebooks, frameworks, MCP servers, tools, and agents, including shadow AI and unapproved coding agents. On top of that inventory it runs RedGraph, an autonomous red-teaming engine that graphs the AI attack surface and runs multi-step attack scenarios the way a real attacker would, chaining tool calls and mapping exploitable surfaces with environmental context.

The third leg is runtime defense: adaptive guardrails calibrated per application or agent that block prompt injection, jailbreaks, and PII or secret leakage in real time, with behavioral-deviation detection that flags when an agent strays from its defined purpose. Pillar claims red-teaming findings and production telemetry continuously recalibrate the guardrails; treat the efficacy claims as vendor marketing, though the research output is independently visible. The company's research team disclosed a critical RCE in n8n (patched in v2.4.0, January 2026), a Week of Sandbox Escapes across Cursor, Codex, Gemini CLI, and Antigravity, and the BadEngram backdoor research on open-weight models (October 2026). Recognition includes Gartner Cool Vendor in AI Software Security 2026, Latio's 2026 AI Security Technical Innovator, and CRN's 10 Hottest AI Security Startups of 2026.

1. Contact the Pillar Security team via pillar.security for platform access (enterprise SaaS).
2. Connect the discovery module to your cloud, code repos, and agent frameworks to build the AI asset inventory.
3. Run RedGraph red-teaming against the discovered surfaces and review the replayable findings.
4. Deploy adaptive runtime guardrails in front of production agents and tune policies from the telemetry feedback loop.