Detectors
Detectors filter input and output data for unwanted content.
You can configure detectors to detect anything from broad content categories to highly specific information.
Detectors can integrate with automated red teaming and adjust detector topics and thresholds to model vulnerability points.
Designed for high throughput applications, detectors are optimized for highly efficient latency and cost.
Use cases
- Filtering for dangerous or illegal content
- Catching personally identifiable information such as SSNs, phone numbers, customer names, etc.
- Flagging jailbreak and adversarial attack attempts
- Preventing off topic or sensitive responses in specific business contexts
- Highlighting likely hallucinations