Lattice: Generative Guardrails for Conversational Agents
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917220876025856 |
|---|---|
| author | Broadhurst, Emily Safi, Tawab Edell, Joseph Ganesh, Vashisht Maamari, Karime |
| author_facet | Broadhurst, Emily Safi, Tawab Edell, Joseph Ganesh, Vashisht Maamari, Karime |
| contents | Conversational AI systems require guardrails to prevent harmful outputs, yet existing approaches use static rules that cannot adapt to new threats or deployment contexts. We introduce Lattice, a framework for self-constructing and continuously improving guardrails. Lattice operates in two stages: construction builds initial guardrails from labeled examples through iterative simulation and optimization; continuous improvement autonomously adapts deployed guardrails through risk assessment, adversarial testing, and consolidation. Evaluated on the ProsocialDialog dataset, Lattice achieves 91% F1 on held-out data, outperforming keyword baselines by 43pp, LlamaGuard by 25pp, and NeMo by 4pp. The continuous improvement stage achieves 7pp F1 improvement on cross-domain data through closed-loop optimization. Our framework shows that effective guardrails can be self-constructed through iterative optimization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_17481 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Lattice: Generative Guardrails for Conversational Agents Broadhurst, Emily Safi, Tawab Edell, Joseph Ganesh, Vashisht Maamari, Karime Artificial Intelligence Conversational AI systems require guardrails to prevent harmful outputs, yet existing approaches use static rules that cannot adapt to new threats or deployment contexts. We introduce Lattice, a framework for self-constructing and continuously improving guardrails. Lattice operates in two stages: construction builds initial guardrails from labeled examples through iterative simulation and optimization; continuous improvement autonomously adapts deployed guardrails through risk assessment, adversarial testing, and consolidation. Evaluated on the ProsocialDialog dataset, Lattice achieves 91% F1 on held-out data, outperforming keyword baselines by 43pp, LlamaGuard by 25pp, and NeMo by 4pp. The continuous improvement stage achieves 7pp F1 improvement on cross-domain data through closed-loop optimization. Our framework shows that effective guardrails can be self-constructed through iterative optimization. |
| title | Lattice: Generative Guardrails for Conversational Agents |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2601.17481 |