// LABS · PLATFORMS, AGENTS, DATA & RESEARCH

Every SAFE SQUARE security service runs on this technology.

We connect collaboration workflows, specialized agents, shared tools, analytics, data, and evaluation into one governed environment — all of it researched and built in-house to make our security services and training better.

01Collaboration platform
02Agents & shared tools
03Analytics & monitoring
04Data · knowledge · evaluation
// 01 · WHAT WE BUILD

One collaboration platform connects
specialized agents, tools, solutions, and data.

Each component is put to work on real engagements and training as it matures — and that maturity is the quality of our services.

01

Human–Agent Collaboration & Orchestration Platform

GOVERNED BY DEFAULT

Coordinates and verifies work across people and multiple agents. Permissions, approvals, audit trails, and quality controls are part of the operating design.

Role & capability registry · task decomposition & handover · human approval for risky ops · execution history & cross-verification
02

Security AI Agents & Shared Tools

VIA TOOL GATEWAY

Specialized SAFE SQUARE agents and shared tools are used side by side by people and approved external agents, connected through a governed Tool Gateway.

Asset & exposure discovery · vulnerability & attack-path analysis · pentest support · AI · OT · IoT · robot evaluation · evidence & reporting
03

Analytics, Assurance & Monitoring Solutions

DISCOVER → ANALYZE → VERIFY → MONITOR → IMPROVE

Discovery, analysis, validation, and continuous monitoring are connected in a single operational workflow.

Attack-surface & leak monitoring · large-scale data analysis · remediation tracking & re-verification · training & CTF platform
04

Security Data, Knowledge & Evaluation Assets

ACCUMULATED SECURITY CAPABILITY

Durable security capability comes from accumulated data, playbooks, and evaluation standards rather than dependence on one model. Client data remains isolated from shared knowledge.

Vulnerabilities · threats · scenarios — methodologies · playbooks · industry baselines — agent evaluation data & benchmarks

* Each part sits at a different stage — R&D, internal use, or already in service delivery — and most are not offered as external products yet.

// 02 · ARCHITECTURE

A model alone is not enough.
We design collaboration, tools, data, and governance as one system.

Models can change as needs and constraints evolve. Tools, data, playbooks, operating history, and human judgment remain available across the ecosystem.

LAYER 1 Users & organizations Security experts · client staff · partners · executives & approvers
LAYER 2 Collaboration & orchestration Task decomposition & assignment · agent registry & capability management · approval · stop · quality gates
LAYER 3 Agents SAFE SQUARE specialized agents · client & partner agents · external general & niche agents
LAYER 4 Tools & execution (Tool Gateway) Asset discovery · scanning · analysis — evidence & monitoring — SIEM · SOAR · ticketing — industrial · IoT · robot tools
LAYER 5 Data, knowledge & evaluation Vulnerabilities · threats · assets — playbooks · policies — project history · evaluation · benchmarks
LAYER 6 Models — swappable Frontier LLMs ⇄ commercial APIs ⇄ local & self-hosted ⇄ task-specialized models

* Shared governance spans identity, permissions, secrets, data isolation, encryption, logging, audit, policy, evaluation, approval, and safety. Deployment can be designed for cloud, hybrid, on-premises, or isolated-network environments.

// OPEN BY DESIGN

The architecture is not limited to SAFE SQUARE agents. It is designed to connect approved external agents and tools in stages through standard interfaces and verified adapters. Every connection is governed by checks for identity, capability, permissions, data access, logging, and quality.

// 03 · CYBER-PHYSICAL

Factories, robots, and smart devices
need more than digital security tools.

From industrial protocol, device, and firmware identification to testbed-based vulnerability assessment and zero-day discovery — purpose-built tools and data for industrial-environment verification.

Targets & tooling
  • OT · ICS · SCADA · PLC · HMI
  • Factories · production lines · industrial networks
  • IoT · smart devices · gateways
  • Firmware · embedded · wireless
  • Humanoids · industrial robots · AMRs
  • Automotive · maritime · transit · logistics
  • Code · binary · update · key analysis
  • Sensor · actuator · control-command verification
Agent deployment principles
  • 01Autonomous attacks are never the default in production environments.
  • 02We start with read, analyze, plan, and evidence work, then expand in stages.
  • 03Risky execution goes through field experts and human approval.
  • 04Results express both cyber risk and physical & operational impact.
// 04 · HOW IT REACHES YOU

Today, through our services and training.
Productization is for later.

This technology is our in-house capability for delivering better services and training. We don't yet offer it as something clients adopt directly.

TODAY

It works inside our services & training

This technology runs inside SAFE SQUARE's security services and training. Clients receive it as faster, sharper delivery.

LATER

Offering the tech itself is a future direction

Making the technology directly usable as products or solutions is open as a direction, but nothing is concretely planned yet.

// 05 · RESEARCH

Questions raised in the field
become Labs research topics.

Labs research starts in the field. Problems met on real engagements become tools, data, and playbooks — and flow back into the field. Security experts and AI agents run this loop together.

AI, LLM & agent security

Attack surfaces of models and agent services, prompt attacks, and data-leak defenses.

Human–AI security operations

Human–agent workflows governed by roles, permissions, approvals, and verification.

Industrial & cyber-physical security

Vulnerability assessment and zero-day discovery across OT, ICS, IoT, firmware, and robots.

Attack & defense knowledge

Field techniques structured into methodologies and industry playbooks.

Agent evaluation & benchmarks

Standards for judging agent capability and safety in security work.

What we learn from real-world verification flows back into the quality of our services and tools. Research collaboration and talent inquiries are welcome.

Collaborate →
// 06 · WHY THIS ARCHITECTURE

Designed for Sovereign Security AI

Our Sovereign Security AI approach explains why models should remain replaceable while tools, data, verification, and operational knowledge stay within the ecosystem.