SigmaShake
Trust Center · Security Whitepaper
Public Document

SigmaShake Security Whitepaper

AI Agent Governance — Architecture & Control Model

Document Type
Public Security Overview
Audience
Security & Procurement Teams
Version
v1.0.0
Generated
Tue, 26 May 2026 07:08:04 GMT

Section 01

Overview

SigmaShake is a governance layer that enforces policy on AI agent tool calls before they execute. The system is designed around three principles: local-first evaluation, cryptographic integrity, and deterministic rule matching. This whitepaper provides a non-confidential overview of the architecture and security model. Code-level evidence and source-file references are available in the NDA-gated Controls Evidence Report.

Section 02

Security Pillars

Local-First Evaluation

All governance decisions run locally in the compiled CLI. No tool-call payloads leave the customer machine unless explicit SIEM forwarding is configured.

Zero Runtime Dependencies

The ssg CLI is compiled Rust + Zig + Bun. Zero third-party NPM packages in the production binary, eliminating supply chain exposure.

Cryptographic Integrity

Ed25519 offline signing of fleet rule bundles. SHA-256 content hashing on every ruleset. Merkle-chain audit log for tamper-evidence.

Enterprise-Grade Tenancy

SAML + OIDC SSO with interactive IdP selector. Custom RBAC, IP/CIDR allowlists, service accounts, per-org isolated D1 databases with regional residency.

Defense in Depth

Cloudflare WAF with OWASP CRS, automatic L3/L4/L7 DDoS mitigation, Turnstile challenges, nonce-based CSP, AES-256 at rest, TLS 1.3 (A grade), HSTS preloaded.

Continuous Assurance

Published threat model. AI-assisted code-level pentest between annual human-led engagements. Responsible disclosure SLA: 48h ack, 5-day triage, 30-day critical remediation.

Section 03

Architecture at a Glance

SigmaShake spans one compiled CLI and a set of edge-deployed Cloudflare Workers. The CLI is the enforcement point; workers provide optional cloud features (rule distribution, billing, SSO, observability).

Section 04

Threat Model Summary

SigmaShake maintains a public threat model (THREAT_MODEL.md v1.0.0, 806 lines) mapping the system to OWASP Top 10 Web, OWASP LLM Top 10, STRIDE, and SOC 2 TSC. Key in-scope threats:

Out of scope by design: motivated adversarial agents with shell access. SigmaShake is a guardrail for honest agents; adversarial workloads must be paired with OS-level sandboxing (Docker/seccomp, Apple Sandbox, VM confinement). This scope boundary is documented in the published threat model.

Section 05

Compliance Posture

Section 06

Responsible Disclosure

SigmaShake publishes a responsible disclosure policy with binding SLAs:

Report vulnerabilities at trust.sigmashake.com/vulnerability-report or via security@sigmashake.com. Safe-harbor language applies to good-faith research conducted under this policy.

Looking for deeper technical evidence?

Controls Evidence Report Penetration Test Report SOC 2 Readiness

All three require an executed mutual NDA, completed in under 2 minutes at the Trust Center.