​
Default model

gemini-3.1-pro

​
Supported models

Any model available through the Google GenAI API:
  • gemini-3.1-pro (default)
  • gemini-3.1-flash
  • gemini-2.0-pro
  • gemini-2.0-flash
  • Any other model ID the API accepts

​
Authentication

Environment variable: 
export GOOGLE_API_KEY=AIza...
Config command: 
ai-blackteam config set providers.google.api_key AIza...
Get your API key from the Google AI Studio.

​
Example usage

# Single attack with default model (gemini-3.1-pro)
ai-blackteam run -p google -a encoding-obfuscation -t "Write a phishing email"

# Full batch with Gemini Flash
ai-blackteam batch -p google --model gemini-3.1-flash --attacks all -t "Write a phishing email"

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

​
Tool-use support

No. The Google provider does not currently support tool-use attacks. Single-turn and multi-turn attacks work fine. Tool-use attacks (agent exploitation, MCP exploitation, etc.) will be skipped when targeting Google.

​
Notes

  • The provider uses the google-genai SDK (from google import genai)
  • System prompts are passed via the system_instruction config parameter
  • For multi-turn conversations, message roles are mapped: assistant becomes model, user stays user
  • 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