Red team
Test the attack path.
Start with an attacker’s access. Test whether an application flaw, exposed identity or unsafe tool call can become a route to critical systems.
The question What can an adversary actually do?
Authorised security assessments
Penetration testing. Built on evidence.
We test where access can cross a boundary in your applications, cloud or AI systems. Your team gets evidence to guide remediation and any agreed retest.
Testing begins only within an agreed, written scope.
A system is only as strong as the boundaries between its parts.
An architectural study of trust boundaries.Across your attack surface
- Web & APIs
- Cloud & IAM
- AI & agent systems
- Internal infrastructure
01 / The threat today
One weakness can become wider access.
An ordinary account reaching another user's data. A cloud identity with more access than its task needs. An AI agent acting on untrusted content. Different openings; the question is what an attacker can reach next.
Test what a foothold can reach.
A weakness matters when it leads to wider access. Genaura tests those connections under written authorisation, giving engineers evidence to remediate and defenders a path to investigate.
Blue team
Use the findings to improve defence.
Turn findings into stronger controls, focused detection improvements and practical remediation. Retest whether the changes stop the original path.
The outcome A clearer path to prevention and detection.
You cannot predict every new exploit. You can test what an attacker could reach.
02 / Assessment scope
Test the systems that connect.
Applications, identities and infrastructure work together. We follow the connections between them, not just a list of individual checks.
Explore assessment servicesThe assessment lens
Who can act?
What can they reach?
Where should the boundary hold?
Assessment surfaces01—04
01Web applications & APIs
Test the assumptions behind authentication, access control and business logic. Follow the path from an ordinary request to an unintended action.
Sessions · Authorisation · Business logic
A question worth testing
Can an ordinary account do something it should not?
02Cloud infrastructure & IAM
Examine how identities, workloads and accounts trust one another. Identify where permissions or configuration turn a foothold into wider access.
Workload identity · Privilege · Isolation
A question worth testing
How far can one compromised identity reach?
03AI & agent systems
Assess what happens when untrusted content reaches a model with access to tools, data and infrastructure. Test the boundary between a response and an action.
Prompt injection · Tool use · Tenant boundaries
A question worth testing
When does untrusted input become a privileged action?
04Internal infrastructure
Evaluate lateral movement and privilege escalation under an agreed assumed-breach scenario. Connect observed attack paths to hardening and detection priorities.
Directory services · Segmentation · Telemetry
A question worth testing
What stands between a foothold and your critical systems?
Scope follows your architecture. Testing follows written authority.
03 / What you receive
An attack path.
A defensible decision.
Finding a weakness is only useful if your team can act on it. Understand how it works, what it puts at risk and what needs to change.
- For security leaders
- Clear impact, explicit limitations and priorities grounded in the tested environment.
- For engineering & defence
- Reproduction steps, remediation guidance and the context needed to validate a change.
Structure preview, not a client report. Contents follow the agreed scope.
04 / Research focus
Where can an AI agent
cross a boundary?
We study where AI systems could let untrusted information become a privileged action. Separately, we are developing AI-assisted security testing methods; that work is still research, not a product.
Explore our research focus
Questions guiding our research
Prompt‑to‑tool boundaries
When does untrusted context become an authorised system action?
Multi-tenant retrieval
Can data cross tenant boundaries through retrieval or shared context?
Agentic cloud identity
What can an agent reach through the permissions it inherits?
Research questions, not published findings.
These questions define the focus, not a claimed result.
05 / Working together
From written scope to retest.
Agree the targets, permissions and stop conditions first. Document what was tested, what was found and what changed after remediation.
Agreed before testing
- Written scope and stop conditions.
- Evidence handling and escalation.
- 01
Define the scope
Agree objectives, target systems, permissions, test windows and stop conditions before testing begins.
OutputWritten scope & rules of engagement
- 02
Test the boundaries
Assess the agreed attack surface, reproduce findings and communicate material risks through the agreed escalation route.
OutputReproducible technical evidence
- 03
Make the findings usable
Connect technical evidence to business impact, remediation priorities and the teams responsible for action.
OutputAssessment report & remediation guidance
- 04
Retest the changes
Revisit the original path within the agreed retest scope. Distinguish resolved findings from open or untested items.
OutputRetest record & remaining limitations
Start an assessment
Start with the question you need answered.
A launch approaching. A new AI integration. A question your last assessment left open. Start with what you need to know; we’ll help define what to test.
We can arrange an NDA before sensitive scoping.
Share scope and timing—not credentials or sensitive data.