auto: 2026-06-06_06:00 changes

1 files changed, 7 insertions, 0 deletions

diff --git a/downloads/newos4_whitepaper.html b/downloads/newos4_whitepaper.html
index 006113a..46c1397 100644
--- a/downloads/newos4_whitepaper.html
+++ b/downloads/newos4_whitepaper.html
@@ -242,6 +242,13 @@
             <li>✅ Predictive resource allocation — task duration/turns prediction from historical data.</li>
             <li>✅ Global World Model — system self-awareness synthesis from all subsystems.</li>
         </ul>
+
+        <h3>Phase 4 — Agent Safety Security Layer <span class="status-done">✅ RELEASED</span></h3>
+        <ul>
+            <li>✅ <strong>Guardrail Layer</strong> — Pre-execution policy engine (8 rules) intercepts AI tool calls matching dangerous patterns (<code>rm -rf</code>, <code>curl|bash</code>, <code>chmod 777</code>, <code>eval</code>, mass-delete, sensitive metadata deletion, external downloads, env overwrites). Policy stored in localStorage, editable via <code>guardrail-add</code> / <code>guardrail-remove</code>.</li>
+            <li>✅ <strong>Confidence Escalation</strong> — AI must declare <code>Confidence: N%</code> in its reasoning before executing destructive actions. Below 90% → manual approval. No confidence declared → manual approval required by default.</li>
+            <li>✅ <strong>Shadow VFS</strong> — <code>shadow &lt;command&gt;</code> snapshots VFS state, runs the command, diffs all filesystem changes. User can <code>shadow-commit</code> or <code>shadow-rollback</code>. AI's <code>run_command</code> tool auto-snapshots and appends shadow diff to every result.</li>
+        </ul>
     </section>
 
     <footer>