About
An evaluation is worth only as much as the independence of whoever produced it. This page says what that means in practice, and who stands behind it.
Independence
An evaluation letter is worth sharing only if your customers can trust it. Three things make that possible:
Deep access, no stake.
We work closely with your agent — its tools, its architecture, its behaviour under attack — but have no stake in the outcome: we don’t build or sell agents, or the products that protect them.
The result is what we measured.
The report lists every threat scenario we ran, how often, and with what result — including the attacks that failed. In a re-run, those show what still holds. Your customers see what was tested, not just what was found.
The method is public.
Anyone can read how a result is produced, and no verdict rests on the agent’s own account or on another model’s judgment of it.
Felix Schwyzer 
(opens in a new tab)
Phaneia was founded by Felix Schwyzer, who has spent more than a decade in cybersecurity. His research has focused on low-level security boundaries such as the one between user mode and kernel, on sandboxing, and on the instrumentation that observes systems from beneath. At CrowdStrike he built core AI systems for detecting and preventing attacks. Phaneia brings these together: it asks whether the boundaries around AI agents hold under attack, observed from where the agent has no access. Named inventor on eight published patents.
Contact
For evaluation enquiries, research correspondence, or press: write to us directly.
Phaneia GmbH · Berlin