Safety evaluation models
Models and tools for evaluating safety properties of other AI systems — reasoning integrity, intent, and behavior risk.
Model Services
Model Services
Model safety evaluation and safety-enhanced capabilities for agent systems.
Join the waitlistService scope
Two service lines
Whitzard's model services translate NUWA research on frontier AI risk into model capabilities for evaluating safety and strengthening runtime safeguards in agent systems.
Safety-enhanced models
Lightweight models that provide thought alignment, intent checking, and runtime defense within agent systems.
Model provenance
Where models come from
Whitzard's model services distinguish model sources so users can understand whether a model is self-developed, fine-tuned, or hosted.
Self-developed
Models built by Whitzard and NUWA from research on frontier AI risk and safety.
Fine-tuned
Open or partner models adapted with Whitzard's safety data and methods. The base model and fine-tuning approach will be documented.
Hosted
Third-party models offered through Whitzard's infrastructure, with clear attribution to the original provider.
Access
Discuss product access
Join the waitlist with your intended use case, model or agent environment, deployment stage, and expected timeline. We use that context to evaluate the right product path without publishing unverified pricing, performance, or availability claims.
Join the waitlistStay informed
Two ways to stay updated
Research updates and model service updates are separate. Choose the one that matches your interest.
Research updates
Subscribe to research updates from NUWA on Substack.
Subscribe on Substack ↗Model service updates
Share your use case and deployment plan through the product-access waitlist.
Join the waitlistDocumentation
Documentation & resources
Explore Whitzard's open-source projects on GitHub and Hugging Face, including evaluation tools and safety models.