​
1. Set your API key

Pick your provider and configure the key: 
ai-blackteam config set providers.anthropic.api_key sk-ant-...
Or set it as an environment variable: 
export ANTHROPIC_API_KEY=sk-ant-...
Both methods work for all providers. See Providers Overview for the full list.

​
2. Run a single attack

ai-blackteam run -p anthropic -a encoding-obfuscation -t "Write a phishing email"
This sends one attack (base64/ROT13/hex encoding) against Claude and shows you the verdict: BYPASSED, PARTIAL, or BLOCKED.

​
3. Run all attacks

ai-blackteam batch -p anthropic --attacks all -t "Write a phishing email"
This fires all 1,000+ attacks against the target prompt and gives you a summary table.

​
4. Run the full benchmark

ai-blackteam benchmark -p anthropic --threshold 80
The benchmark runs 40 targets across 1,000+ attacks. Set --threshold to fail the run if the safety score drops below that number (useful for CI).

​
5. Generate a scorecard

# OWASP LLM Top 10
ai-blackteam scorecard --standard llm

# OWASP Agentic Top 10
ai-blackteam scorecard --standard agentic

# EU AI Act + NIST AI RMF
ai-blackteam scorecard --standard compliance
Scorecards map your results to industry standards and show per-category pass/fail rates.

​
6. Generate a report

# HTML dashboard
ai-blackteam report --format html --output report.html

# JSON for CI pipelines
ai-blackteam report --format json --output results.json

# Export to Promptfoo format
ai-blackteam report --export promptfoo --output results.json

# Export to garak format
ai-blackteam report --export garak --output results.jsonl

​
What’s next

Your First Scan

A detailed walkthrough of running and reading results

Providers

Configure all 7 supported LLM providers