🛡️ Generative Shield

Plattform-Cockpit · checking

📊 Live-Statistiken

AUTO-REFRESH 15s
—
Total Calls
—
Blocked · Jailbreak
—
Blocked · PII
—
Masked · PII
—
Passed
—
Uptime (min)
—
L3 · Flagged Outputs
—
Auto-Skills (Forge)

🏛️ 3-Layer Architektur

3 / 3 LIVE

Defense-in-Depth gegen Prompt-Injection, PII-Leak, Toxicity, Bias und Jailbreak.

L1
Datenschutz · Presidio
DSGVO Art. 25 · PII-Mask · 8 Entity-Types
ACTIVE
L2
Policy · Pattern-Filter
12 Jailbreak-Patterns DE+EN · Sensitivity 70%
ACTIVE
L3
Safety-Net · llm-guard
Toxicity, Bias, Secret-Leak · Pattern (BERT W70)
ACTIVE

🎚️ Cybersecurity Sensitivity

ACTIVE

Jailbreak-Detection-Schwellwert. 0% = permissiv, 100% = sehr streng.

Sensitivity
L2 Pattern-Threshold
70%
PII Block-Threshold
L1 Presidio Confidence
0.70
L3 Output Threshold
Score über dem geblockt wird
70

🤖 Self-Improving Agents

W68 LIVE

KI-Agents legen autonom Skills, Knowledge-Bases, Prompts und n8n-Workflows an.

—
Tools
—
Knowledge-Bases
—
Prompts
—
Workflows

Forge-Endpoints: /forge/owui/tool · /forge/owui/knowledge · /forge/owui/prompt · /forge/n8n/workflow

📁 Filesystem-MCP

ACTIVE

Tenant-sandboxed File-Access für KI-Agents. Path-traversal-protected.

—
Operations
10MB
Max File Size

Operations: read · write · list · mkdir · delete · search

🧪 Live-Test gegen Shield

REGEX + L1+L2

Teste einen Prompt sofort gegen alle Layer ohne LLM-Call. Spart Kosten.

Klick "Test" um zu starten...

📜 Aktivitäts-Feed (Forge-Audit)

LIVE

Alle vom Skill-Forge erzeugten Skills · Audit-Trail für EU-AI-Act.

load
lade Aktivitäten