DynaGuard: A Dynamic Guardian Model With User-Defined Policies
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912632312692736 |
|---|---|
| author | Hoover, Monte Baherwani, Vatsal Jain, Neel Saifullah, Khalid Vincent, Joseph Jain, Chirag Rad, Melissa Kazemi Bruss, C. Bayan Panda, Ashwinee Goldstein, Tom |
| author_facet | Hoover, Monte Baherwani, Vatsal Jain, Neel Saifullah, Khalid Vincent, Joseph Jain, Chirag Rad, Melissa Kazemi Bruss, C. Bayan Panda, Ashwinee Goldstein, Tom |
| contents | Guardian models play a crucial role in ensuring the safety and ethical behavior of user-facing AI applications by enforcing guardrails and detecting harmful content. While standard guardian models are limited to predefined, static harm categories, we introduce DynaGuard, a suite of dynamic guardian models offering novel flexibility by evaluating text based on user-defined policies, and DynaBench, a dataset for training and evaluating dynamic guardian models. Our models provide both rapid detection of policy violations and a chain-of-thought reasoning option that articulate and justify model outputs. Critically, DynaGuard not only surpasses static models in detection accuracy on traditional safety categories, but is competitive with frontier reasoning models on free-form policy violations, all in a fraction of the time. This makes DynaGuard an critical tool for language model guardrails. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_02563 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DynaGuard: A Dynamic Guardian Model With User-Defined Policies Hoover, Monte Baherwani, Vatsal Jain, Neel Saifullah, Khalid Vincent, Joseph Jain, Chirag Rad, Melissa Kazemi Bruss, C. Bayan Panda, Ashwinee Goldstein, Tom Machine Learning Computation and Language Guardian models play a crucial role in ensuring the safety and ethical behavior of user-facing AI applications by enforcing guardrails and detecting harmful content. While standard guardian models are limited to predefined, static harm categories, we introduce DynaGuard, a suite of dynamic guardian models offering novel flexibility by evaluating text based on user-defined policies, and DynaBench, a dataset for training and evaluating dynamic guardian models. Our models provide both rapid detection of policy violations and a chain-of-thought reasoning option that articulate and justify model outputs. Critically, DynaGuard not only surpasses static models in detection accuracy on traditional safety categories, but is competitive with frontier reasoning models on free-form policy violations, all in a fraction of the time. This makes DynaGuard an critical tool for language model guardrails. |
| title | DynaGuard: A Dynamic Guardian Model With User-Defined Policies |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2509.02563 |