[ Safety ]

Responsible deployment.

Red-teaming, injection defenses, evaluation cards, and deployment guardrails.

Focus areas

Versioned red-team packs on every candidate build.

Filters and monitoring on top of model refusals.

5 papers in this area

Compact safety scorecard for Origin, NextGen, and Quark—companion to the full Safety Research report.

IdonAI’s safety framework for Origin, NextGen, and Quark — threat modeling, red-teaming, evaluations, product guardrails, and limitations.

How we lowered benign over-refusal toward single digits while holding jailbreak metrics steady.

Indirect prompt injection threat model, evaluations, and mitigations for Quark’s document-aware workflows.

First standardized jailbreak and misuse suite used as a hard gate before Origin promotions.

[ Research ]

Browse every area, or email the research team about collaborations, datasets, and eval access.