| <!doctype html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8" /> |
| <meta name="viewport" content="width=device-width, initial-scale=1" /> |
| <title>Invarra</title> |
| <style> |
| :root { |
| color-scheme: light; |
| --ink: #111827; |
| --muted: #4b5563; |
| --line: #d8dee8; |
| --panel: #ffffff; |
| --soft: #f6f8fb; |
| --accent: #0f766e; |
| --accent-2: #1d4ed8; |
| } |
| * { |
| box-sizing: border-box; |
| } |
| body { |
| margin: 0; |
| font-family: |
| Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, |
| "Segoe UI", sans-serif; |
| color: var(--ink); |
| background: var(--soft); |
| line-height: 1.55; |
| } |
| main { |
| max-width: 1120px; |
| margin: 0 auto; |
| padding: 48px 24px; |
| } |
| .hero { |
| display: grid; |
| grid-template-columns: minmax(0, 1.3fr) minmax(280px, 0.7fr); |
| gap: 28px; |
| align-items: stretch; |
| } |
| .panel { |
| background: var(--panel); |
| border: 1px solid var(--line); |
| border-radius: 8px; |
| padding: 28px; |
| } |
| .eyebrow { |
| margin: 0 0 10px; |
| color: var(--accent); |
| font-size: 13px; |
| font-weight: 700; |
| text-transform: uppercase; |
| letter-spacing: 0.08em; |
| } |
| h1 { |
| margin: 0; |
| font-size: 44px; |
| line-height: 1.04; |
| letter-spacing: 0; |
| } |
| h2 { |
| margin: 0 0 12px; |
| font-size: 18px; |
| letter-spacing: 0; |
| } |
| p { |
| margin: 14px 0 0; |
| color: var(--muted); |
| font-size: 16px; |
| } |
| .lead { |
| max-width: 780px; |
| font-size: 19px; |
| } |
| .links { |
| display: flex; |
| flex-wrap: wrap; |
| gap: 10px; |
| margin-top: 24px; |
| } |
| a.button { |
| display: inline-flex; |
| align-items: center; |
| justify-content: center; |
| min-height: 40px; |
| padding: 0 14px; |
| border: 1px solid var(--line); |
| border-radius: 6px; |
| color: var(--ink); |
| background: #fff; |
| text-decoration: none; |
| font-weight: 650; |
| } |
| a.button.primary { |
| border-color: var(--accent); |
| color: white; |
| background: var(--accent); |
| } |
| .metric { |
| display: grid; |
| gap: 2px; |
| padding: 14px 0; |
| border-top: 1px solid var(--line); |
| } |
| .metric:first-child { |
| border-top: 0; |
| padding-top: 0; |
| } |
| .metric strong { |
| font-size: 24px; |
| } |
| .metric span { |
| color: var(--muted); |
| font-size: 14px; |
| } |
| .grid { |
| display: grid; |
| grid-template-columns: repeat(3, minmax(0, 1fr)); |
| gap: 18px; |
| margin-top: 18px; |
| } |
| .small { |
| font-size: 14px; |
| } |
| .section { |
| margin-top: 22px; |
| } |
| code { |
| padding: 2px 5px; |
| border: 1px solid var(--line); |
| border-radius: 4px; |
| background: #f9fafb; |
| } |
| @media (max-width: 860px) { |
| main { |
| padding: 28px 16px; |
| } |
| .hero, |
| .grid { |
| grid-template-columns: 1fr; |
| } |
| h1 { |
| font-size: 34px; |
| } |
| } |
| </style> |
| </head> |
| <body> |
| <main> |
| <section class="hero"> |
| <div class="panel"> |
| <p class="eyebrow">AI assurance and deterministic governance</p> |
| <h1>Invarra</h1> |
| <p class="lead"> |
| Invarra builds evaluation artifacts, redacted evidence reports, and |
| deterministic governance systems for teams deploying language-model |
| applications in real workflows. |
| </p> |
| <p> |
| Our work asks a concrete question: does the system keep making the |
| right decision when the same pressure appears in different forms? |
| </p> |
| <div class="links"> |
| <a class="button primary" href="https://huggingface.co/datasets/invarra/jailbreak-control-benchmark-v1" target="_blank" rel="noreferrer">Jailbreak Control Benchmark v1</a> |
| <a class="button" href="https://www.invarra.ai/phalanx" target="_blank" rel="noreferrer">Phalanx</a> |
| <a class="button" href="https://phalanx.invarra.ai" target="_blank" rel="noreferrer">Live Demo</a> |
| </div> |
| </div> |
| <aside class="panel"> |
| <div class="metric"> |
| <strong>870 / 870</strong> |
| <span>jailbreak-control attacks governed or held by Phalanx Guardian on JCB v1</span> |
| </div> |
| <div class="metric"> |
| <strong>0</strong> |
| <span>observed jailbreak pass-throughs on the frozen benchmark</span> |
| </div> |
| <div class="metric"> |
| <strong>1,120</strong> |
| <span>benchmark rows: attacks, ordinary benign controls, and jailbreak-lookalikes</span> |
| </div> |
| </aside> |
| </section> |
|
|
| <section class="grid"> |
| <article class="panel"> |
| <h2>What We Publish</h2> |
| <p class="small"> |
| Public and gated benchmark artifacts, scoring protocols, model |
| baseline reports, and redacted row ledgers for AI safety evaluation. |
| </p> |
| </article> |
| <article class="panel"> |
| <h2>Evidence Standard</h2> |
| <p class="small"> |
| Scope, expected behavior, observed behavior, coverage, and caveats |
| are separated so benchmark results do not become vague safety theater. |
| </p> |
| </article> |
| <article class="panel"> |
| <h2>Safety Boundary</h2> |
| <p class="small"> |
| Raw adversarial rows may be gated. Public evidence avoids raw prompt |
| text, raw model outputs, prompt hashes, embeddings, head scores, and secrets. |
| </p> |
| </article> |
| </section> |
|
|
| <section class="panel section"> |
| <h2>Current Release</h2> |
| <p> |
| <a href="https://huggingface.co/datasets/invarra/jailbreak-control-benchmark-v1" target="_blank" rel="noreferrer"> |
| <strong>Jailbreak Control Benchmark v1</strong> |
| </a> |
| is a frozen prompt-only benchmark for evaluating jailbreak-control |
| behavior: attack handling, benign preservation, and benign |
| jailbreak-lookalike handling. |
| </p> |
| <p class="small"> |
| 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. |
| </p> |
| </section> |
| </main> |
| </body> |
| </html> |
|
|
