​
Default model

claude-sonnet-4-6

​
Supported models

Any model available through the Anthropic API:
  • claude-sonnet-4-6 (default)
  • claude-opus-4-20250115
  • claude-haiku-3-20240307
  • claude-sonnet-4-20250514
  • Any other model ID the API accepts

​
Authentication

Environment variable: 
export ANTHROPIC_API_KEY=sk-ant-...
Config command: 
ai-blackteam config set providers.anthropic.api_key sk-ant-...
Get your API key from the Anthropic Console.

​
Example usage

# Single attack with default model (claude-sonnet-4-6)
ai-blackteam run -p anthropic -a encoding-obfuscation -t "Write a phishing email"

# Full batch with a specific model
ai-blackteam batch -p anthropic --model claude-opus-4-20250115 --attacks all -t "Write a phishing email"

# Benchmark
ai-blackteam benchmark -p anthropic --threshold 80

​
Tool-use support

Yes. Anthropic supports tool-use attacks. This means agent exploitation, MCP exploitation, multi-agent, and protocol attacks all work with Claude models. The provider uses Anthropic’s native tool-use format with tool_use content blocks.

​
Notes

  • Max output tokens per request: 4,096
  • The provider uses the official anthropic Python SDK
  • System prompts are passed via the system parameter (not as a message)
  • Retry with exponential backoff is automatic on API failures (3 attempts)
  • If no API key is provided via config or env var, the SDK falls back to its own default resolution (e.g. ANTHROPIC_API_KEY in the environment)