Skip to content
CLI · CI/CD ready

Secure your LLMs.
Prove your compliance.

Shield LLM runs automated security tests on any AI chatbot: prompt injection, data extraction, jailbreaks. Security grade, detailed vulnerabilities, audit-ready PDF report — from a single command.

Automated red teamingOWASP LLM Top 10LLM-as-JudgePDF export
Start scanning →See how it works
Works with any chatbot endpoint
Runs locally
Audit-ready PDF reports
PROMPT_INJECTION_BLOCKED
SYSTEM_PROMPT_SECURED
MODEL_DOS_DETECTED
OUTPUT_SANITIZED
Act I · The reality

Your AI chatbots are already under attack.
Most are vulnerable.

Production LLMs pass classic web audits. They fail against LLM-specific attacks: prompt injection, data extraction, guardrail bypass.

0%
of production chatbots fail basic OWASP LLM tests
OWASP · 2025
0 min
is the average time to extract sensitive data from an unprotected LLM
Shield Labs
0%
of companies have no defense specific to their LLM deployments
Gartner · 2025
0×
increase in attacks targeting generative AI since 2024
IBM X-Force

The attacks a classic scanner never sees

LLMs introduce a completely new attack surface. Your WAF, your DAST, your annual pentest: none of them are equipped for it.

  • LLM01Prompt injection that hijacks model behavior
  • LLM02Sensitive information leakage in responses
  • LLM06Excessive agency (unauthorized actions)
  • LLM07System prompt extraction via social engineering
  • LLM09Disinformation and exploitable hallucinations
Act II · The solution

Three pillars. One tool.

Shield LLM combines automated red teaming, AI-judge analysis and regulatory compliance in a CLI you install in one command.

PILLAR 01

Automated red teaming

Full coverage of the OWASP LLM Top 10: prompt injection, system extraction, excessive agency, multi-turn jailbreak, supply chain. Single-turn and multi-turn attacks.

OWASP LLM TOP 10 · CRESCENDO
PILLAR 02

LLM-as-Judge

AI-driven analysis goes beyond regex. A judge model evaluates each response in context to catch nuanced vulnerabilities that pattern matching misses.

3 LAYERS · CONTEXT + RULES + JUDGE
PILLAR 03

EU AI Act compliance

Automatic scoring aligned with articles 5, 9, 10, 13, 14 and 15 of the regulation. Audit-ready PDF report, actionable remediations, timestamped evidence.

7 REQUIREMENTS · SIGNED PDF EXPORT
shield-llm · real-time scan
TARGETsupport-bot.acme.com
F
23/100
Security grade · immediate action required
PROMPT INJECTION
85%
DATA LEAKAGE
72%
OUTPUT HANDLING
48%
SUPPLY CHAIN
15%
4 critical · 6 high · 3 medium detected
Act III · Regulatory framework

EU AI Act. Audit-ready.

Your LLM falls under the European AI regulation. The obligations are concrete: robustness, transparency, human oversight, data governance. Shield gives you the technical scoring and the evidence.

Articles covered
7 of 7 high-risk system requirements
Enforcement
December 2, 2027 · high-risk systems
Assess my compliance →

Compliance assessment · support-bot.acme

Compliant
Robustness and resilience
Art. 15
92%
Transparency
Art. 13
88%
Data governance
Art. 10
85%
Human oversight
Art. 14
90%
Risk management
Art. 9
87%
Accuracy
Art. 15
91%
Prohibited practices
Art. 5
95%
Local-first architecture

Your prompts never leave your environment.

Shield LLM runs as a CLI from your environment, calling your chatbot endpoint directly. No proxy, no MITM, no credentials to share.

Sent to our servers: your chatbot's replies (to score the scan, never to train models) and the report summary for your dashboard.

  • 100% local execution from your environment
  • Your chatbot credentials stay in your config, never sent to us
  • Data hosted in Europe · encryption at rest
  • Signed PDF export ready for your internal audits

Zero interception

Prompts go straight to your chatbot endpoint. No relay, no proxy.

Credentials stay local

Your shield.config.json lives on your machine. We never see your endpoint secrets.

Isolated storage

Scan data is scoped to your environment. Nothing persists without your action.

Verifiable output

Every result includes evidence, confidence and OWASP mapping. Native PDF export.

FAQ

Frequently asked questions

Everything you need to know about Shield LLM security testing.

What happens during a scan?
Shield LLM automatically sends OWASP LLM Top 10 attack prompts to the target chatbot endpoint. Each response is analyzed by our 3-layer engine (context + rules + LLM-as-Judge). You get an A–F grade, the vulnerability breakdown, and an audit-ready PDF report.
Is my data sent to a server?
Shield LLM is self-hosted: there is no Shield server. Attack prompts go straight to your chatbot, and the replies are scored by the instance you run on your own infrastructure, which also stores the history shown in your dashboard. The only outbound call is the AI judge, to the LLM provider you configure with your own API key — replies are used solely to score the scan, never to train models.
Does it work with any chatbot?
Yes, as long as it is reachable over HTTP. Point the CLI at your endpoint — OpenAI-compatible APIs, Claude, Gemini, Mistral, or any custom chatbot — and shield.config.json handles authentication, headers and request/response mapping. SSE streaming is auto-detected. No model-specific configuration required.
How long does a scan take?
Mostly it depends on your chatbot, not on Shield: attacks run 4 at a time by default, so your endpoint's own latency and rate limits set the pace. A quick scan takes well under a minute. A full scan, which adds the multi-turn attacks, takes several minutes: we measured just under 6 minutes for the attack run against a public test target, and a rate-limiting endpoint takes longer still because the scanner automatically backs off instead of hammering it. Raise --concurrency to trade politeness for speed.
How is the security grade calculated?
Weighted Impact Scoring: every finding carries a severity weight (CRITICAL 10, HIGH 5, MEDIUM 2, LOW 1), and the score is 100 × (1 − observed penalty / total weight) — so one critical finding costs far more than a handful of low ones. Grades map to A (≥90), B (≥75), C (≥60), D (≥40), F (<40). The calculation details are documented in the PDF report.
Does Shield LLM really cover the EU AI Act?
Shield covers the 7 technical requirements for high-risk AI systems (articles 5, 9, 10, 13, 14, 15). Automatic scoring produces timestamped evidence you can attach to your compliance file. Full legal assessment remains your DPO's responsibility.

Secure your AI before the attackers.

Install the Shield LLM CLI and run your first scan in minutes. No SDK, no code changes.

Shield LLM — Automated Red Teaming for AI Chatbots