Glossary term
Glossary term
Safety and Alignment
Controllability refers to how well AI behavior can be guided or constrained through boundaries on responses, tool usage, and tone, ensuring alignment with business rules, compliance standards, and safety in enterprise settings.
Nvidia NeMo Guardrails provides programmable controllability for LLM applications.
AWS Bedrock Guardrails enforces topic, content, and PII filters for Anthropic and other foundation models.
Anthropic's Constitutional AI gives Claude controllability through explicit principles encoded in training.