
Run threat research and AI red teaming
in an isolated environment
Research and AI Red Team assessments take place in an isolated environment.
Outbound transfers are blocked by default. Exporting research results requires approval from the person responsible.
Deployment consultationWe’ll confirm the scope and deployment environment during the consultation. Deployment guide
Research outputs
require approval before export
Each researcher has a separate Kubernetes namespace. A default-deny policy blocks outbound traffic.
Exporting reports or code requires approval from the responsible person and a digital signature.
- The policy denies outbound transfers by default
- Record the file hash and approver for every export
- Separate namespaces for each researcher
lab-ns/target-01In progress
Policy: egress-deny10:41
Awaiting approval from the person responsible10:55
We assess safeguards for
AI models and agents
We use public tools such as Garak and PyRIT and scenarios developed in-house to assess defenses against prompt injection, jailbreaks, and tool manipulation. Researchers maintain Korean-language test cases that English datasets may miss. The report documents confirmed findings and proposes guardrail rules.
- Test cases for direct and indirect prompt injection and tool manipulation
- Korean-language security test cases
- Confirmed findings inform draft guardrail rules
Check the input source against the policyRecords
Check whether approval is requiredReview
Check execution logs against the approved scopeRecords
Who did what, and when
is all recorded
The signed audit log records command execution, access to systems under review, and requests to export files.
Review this record to verify that the research stayed within the approved scope.
- Audit logs for each command
- Session screen recording
- Generate reports for audits
Key features
Isolated research environment
Isolate researcher workspaces with namespaces and NetworkPolicy
Red Team operations
Run exercises within the agreed scope and scenarios
AI Red Teaming
Automated evaluation of LLM and agent safeguards
Vulnerability research
We notify the vendor first about newly discovered vulnerabilities
Audit log
Signed records for each command
Outbound transfer controls
Only approved and signed files can be exported
How it works
Scope approval
Agree in writing on the research purpose and scope
Environment setup
Prepare an isolated workspace and a replica of the system under review
Research
Security research and testing
Review
Review results and approve exports
Apply
Draft Sigma and YARA detection rules and prepare reports
Integration tools and methods
Integrations use APIs and syslog. We’ll confirm what each tool supports based on its version and your operating environment.
Deployment options
| Method | Description | Recommended |
|---|---|---|
| On-premises and isolated networks | The platform runs in your data center. Offline packages deliver detection rule and model updates so the platform can operate without an external connection. | Finance, public sector, and defense |
| Dedicated cloud | A dedicated environment serves one customer in a South Korean cloud region. | Enterprises and healthcare |
| Managed | In this MDR option, Genox handles security monitoring. We’ll confirm the scope and response procedures during the consultation. | Small and medium-sized businesses and mid-market companies |
| MSSP partners | The security service provider operates separate tenants for each customer. | Security service providers |
Frequently asked questions
Can customers ask about using Research Lab?
Organizations with an in-house red team can discuss availability, scope, and terms for an isolated research environment during a deployment consultation.
How are newly discovered vulnerabilities handled?
We follow responsible disclosure: we notify the vendor first and publish after a patch is available.

Talk to us about Research Lab
Tell us about your operating environment, and we’ll explain what the demo and proof of concept can cover.