What We Publish
Public and gated benchmark artifacts, scoring protocols, model baseline reports, and redacted row ledgers for AI safety evaluation.
AI assurance and deterministic governance
Invarra
Invarra builds evaluation artifacts, redacted evidence reports, and deterministic governance systems for teams deploying language-model applications in real workflows.
Our work asks a concrete question: does the system keep making the right decision when the same pressure appears in different forms?
Evidence Standard
Scope, expected behavior, observed behavior, coverage, and caveats are separated so benchmark results do not become vague safety theater.
Safety Boundary
Raw adversarial rows may be gated. Public evidence avoids raw prompt text, raw model outputs, prompt hashes, embeddings, head scores, and secrets.
Current Release
Jailbreak Control Benchmark v1 is a frozen prompt-only benchmark for evaluating jailbreak-control behavior: attack handling, benign preservation, and benign jailbreak-lookalike handling.
A result on this benchmark is not a claim of universal jailbreak immunity, broad content moderation, RAG-injection coverage, tool-safety coverage, multimodal security, or infrastructure security.