Custom Model Red Teaming¶
The Custom Model collector red teams a single model over its HTTP inference endpoint, without onboarding a full cloud account. The flow works for any LLM exposed over HTTP. Pick your serving platform to get started.
The collector you pick depends on where the model is served. Cloud-managed models use the cloud SDK collector. Self-hosted engines use the Ollama or OpenAI collector. A hosted API behind an OpenAI-compatible interface uses the OpenAI interface collector.
Secret tokens for custom models
Managed Bedrock onboarding uses a Bedrock API key. Self-hosted targets such as NVIDIA Triton and vLLM have no default secret token, so leave the token empty unless your endpoint sits behind an auth proxy or was started with an API key.
Prefer full-account coverage?
To scan every model in a cloud account automatically, use the AWS AI/ML Onboard, Azure AI/ML Onboard, or GCP AI/ML Onboard flows instead.