Grey box
Signed in, like a real user. Or a stolen account.
- Public APIs
- Access control across roles
- Prompt injection and LLM leaks
Pentesting · Munich
Break it before they do.
Pentests for AI startups that ship fast and have no security team yet.
Signed scope first. Retest included.
N2-03 · API · Grey boxHigh
One tenant can read another tenant's chat history
GET /api/v1/conversations?tenant=acme-b Authorization: Bearer <token for acme-a> 200 OK [ 42 conversations ]
ImpactData exposure
FixBind tenant to token
RetestFixed
About
Security, without the security team.
We're offensive security and AI engineers. We test what you actually run: your API, your roles, your LLM features. You get findings your engineers can fix, and a retest once they have.
- Written authorization first
- Findings engineers can act on
- Retest included
Services
Two ways in.
Pick how much we know going in.
Black box
Nothing but your domain, like an outside attacker.
- Exposed attack surface
- Sign-up, login and reset flows
- Public LLM features
People
The team.
Max Ngo
Offensive Security Engineer, NAB
Ex Canva · BSc, Swinburne
Duc Nguyen
AI Engineer
Ex Celonis · MSc, TUM
Blog
Field notes.
LLM security
LLM features: what we check first
Prompt injection gets the headlines. The leaks come from the plumbing behind the chat box.
APIs
Multi-tenant APIs: five common mistakes
IDs in URLs, trusted headers, unchecked bulk endpoints.
Process
Your first pentest: how to prepare
Test accounts, a signed scope and who to call when we find something.
Find it first.
A 30-minute call to agree on scope. Nothing starts without your sign-off.