File size: 6,700 Bytes
5eb5274 30ba30e 5eb5274 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 | <!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>
|