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>