OUTIS
AI RED TEAM
Outis — Greek for “Nobody.”
We attack AI systems the way a motivated adversary would, then hand back a measured, evidence-backed account of what actually leaks, with the limits stated plainly.
Read Case Study 01Method
Numbers you can trust
Pre-registered
The metric, the win rule and the number of trials are written down and frozen before any reportable data exists.
Chain of custody
Isolated single-use runners, hash-locked evidence, and a human sign-off before any result counts.
Caveats attached
Every figure ships with its scope, so it can't be read as more than it is.
What we test
Mapped to the OWASP LLM Top 10
LLM01Prompt injection, direct and hidden in content
LLM02Sensitive information disclosure
LLM05Improper output handling
LLM06Excessive agency and tool abuse
LLM07System prompt leakage
LLM08Retrieval (RAG) weaknesses
Testing is done only on systems we own or are authorized in writing to test.
The name
In the Odyssey, Odysseus tells the Cyclops his name is Outis, “Nobody,” and that is how he gets out. A good red team works the same way: unnoticed until the report.