Vigil Security Operations Center

VigilSOC Platform

The Open Source AI Security Operations Center

Primary user base: SOC analysts, threat hunters, and MSSP operators using VigilSOC for automated telemetry ingest, LogLM zero-shot triage, and machine-speed mitigation.

Community-contributed modules: Automated adversary simulation playbooks, synthetic attack generation, and continuous attack-surface probing to stress-test detection boundaries.

Continuous collaborative loop: Real-time feedback engine matching offensive Red simulation against Blue detection DAGs to validate and self-correct playbooks automatically.

Personal Perspective (May 2026 Presentation: "Agentic SOC Operations: OSS AI Tooling"): Frame VigilSOC fundamentally as a Blue Team Defender Tool operating with an explicit, unyielding assumption of Human-IN-the-Loop (HITL) validation at every execution step before taking action.

Personal Perspective (June 2026 Presentation: "The Agentic Battlefield: Red vs Blue Agents"): Noted heavy community adoption of red team tooling while shifting to Human-ON-the-Loop (HOTL) supervisory principles as humans still couldn't keep up with triage.

Personal Perspective (Hacker Summer Camp Takeaways): Focus shifted to full autonomy with strict guardrails, confidence scores, and automatic remediation while incorporating continuous feedback loops for False Positives (FP) and False Negatives (FN).

• 🚫 Static 2,000-Word Prompt Bloat: Giant system prompts packed with rigid "always/never" rules that caused models to hallucinate or freeze.
• 🚫 Rigid Linear Pipelines: Hardcoded step-by-step scripts hiding iteration loops and forcing uniform execution across all incident types.
• 🚫 Opaque Black-Box LLM Wrappers: Hidden prompt text and un-auditable fallback chains impossible for SOC analysts to inspect or verify.
• 🚫 Infinite Loop Computations: Agents blindly chasing dead-end threat branches indefinitely without cost or compute limits.

• 🎯 Dynamic Selection Functions: Token conditioning focusing model attention on precise task coordinates rather than static prompt bloat.
• 🔄 Event-Driven DAG Engine: StackStorm-inspired Directed Acyclic Graph orchestrator for non-linear, stateful threat hunting.
• 🛠️ 30+ Modular MCP Connectors: Decoupled Model Context Protocol tool interfaces (VT, AbuseIPDB, Shodan) executed on-demand.
• 🛡️ Code-Level Pivoting & Early Kill Triggers: First-class pivots (Netflow ➔ Process ➔ Audit Logs) and early branch abandonment to preserve compute.

Version-controlled governance documents defining non-negotiable boundaries, intent, and schema contracts:

• 📜 principles.md: Local-first privacy, zero black-box wrappers, open-source Apache 2.0.
• 🎯 intent.md: Declarative agent goals, hunt boundaries, and threat hunter intuition in code.
• 🏷️ versioning.md: Strict semantic release criteria and API contract stability guarantees.

Concrete runtime execution engine implemented directly in Vigil-SOC/vigil:

• 🔄 Event-Driven DAG: StackStorm-inspired non-linear workflow orchestrator with state tracking.
• 🛡️ Circuit Breakers: First-class domain pivots (Netflow ➔ Process ➔ Audit Logs) & early kill triggers.
• 🔌 30+ MCP Connectors: Standardized Model Context Protocol tool interfaces.

Autonomous factory generating code upstream & syncing to the public repo:

• 🤖 Autonomous PR Authoring: Backlog grooming, unit test generation, and live validation.
• ⚡ Internal Cursor Factory Sync: Proprietary Cursor factory rules, custom scratch scripts, and prompt orchestrators currently being audited for public sync to Vigil-SOC/vigil.

# Vigil SOC — security.txt
# Standard: RFC 9116 (https://www.rfc-editor.org/rfc/rfc9116)
# Canonical location: https://vigilsoc.org/.well-known/security.txt
#
# To report a vulnerability, please use GitHub private vulnerability reporting
# (preferred) or the security email below. Do NOT open public GitHub issues.

Contact: https://github.com/Vigil-SOC/vigil/security/advisories/new
Contact: mailto:[email protected]
Preferred-Languages: en
Policy: https://vigilsoc.org/security
Canonical: https://vigilsoc.org/.well-known/security.txt

Task: Publishing a comprehensive, cryptographically signed Software Bill of Materials (SBOM) with standardized in-repo manifests (CycloneDX / SPDX).
Impact: Provides enterprise consumers with verifiable build attestations, provenance tracking, and automated transitive dependency validation.

Task: Penetration testing and threat modeling targeting internal event-driven DAG execution hooks, multi-agent IPC channels, and external API connectors.
Impact: Prevents malicious webhooks or manipulated payload data from hijacking DAG execution flow or breaching sandbox isolation.

Task: Adversarial red teaming directed at the configured Inference & Model providers (local foundation models, self-hosted LLMs, or cloud API endpoints), evaluating prompt-injection resilience, token manipulation, and model provider integrity.
Impact: Ensures adversarial telemetry or manipulated inputs cannot blind inference encoders, bypass safety guardrails, or trick downstream triage swarms regardless of the underlying model provider.

Task: Active review and refactoring of 30+ Model Context Protocol (MCP) connectors for compliance with latest MCP spec updates.
Impact: Enforces standardized state persistence rules and fine-grained authentication/authorization primitives across all external tool integrations.

Domain: Management Systems.
Focus: Org-wide policy, continuous improvement, & ethical AI governance controls across life cycles.

Domain: Risk Management.
Focus: 4 Core Functions (Govern, Map, Measure, Manage) for trustworthy, bias-aware AI operations.

Domain: Application Security.
Focus: Tactical vulnerabilities (Prompt Injection, Data Poisoning, Insecure Output, Supply Chain).

Domain: Legal Compliance.
Focus: Risk classification (Unacceptable, High, GPAI), requiring conformity assessments & audits.

Domain: Adversary Tactics.
Focus: Threat matrix mapping real-world ML attacks (Model Evasion, Weight Theft, Poisoning).

Domain: Cloud AI Services.
Focus: Security assurance, data privacy guarantees, and tenant isolation for cloud AI deployments.

Connecting AI & cybersecurity leaders across academia and industry. Live Bay Area events hosted at #SVAI in Menlo Park:

Free practitioner-led virtual webinars, interactive workshops, and education on AI security governance & defense:

Join open community office hours with VigilSOC maintainers & threat researchers. Discuss agent DAG workflows, custom MCP tool connectors, playbooks, and real-world deployment telemetry.