Office Hours: July 8, 2026

[annebdh]

Our next office hours are happening Wednesday, July 8, 2026 at 1630–1700 UTC!

Learn how T&S peers are using AI for safety workflows, ask for support in implementing open safety models, and connect directly with RMC model partners. We'll also discuss how you can contribute to ongoing projects that the RMC is working on!

Discord event
Google Meet
#roost-model-community

Proposed Agenda

• Introduction / Updates
• Recap of what the ROOST Model Community is and the purpose of office hours
• The constitutional approach to safety guardrails -- what it is, tradeoffs, experiments
• Open discussion

Reminder

Office hours for July 22nd are cancelled due to the number of people at TrustCon. If you're there, come say hello!

[cassidyjames]

Notes

Attendees: @annebdh, @julietshen, Selena, @cassidyjames, Nav

• Constitutional approach to safety guardrails
  • What is it: create a dataset that has high coverage (a “constitution”) within specific harm categories; wider coverage, deeper depth of training data
  • AstroWare released HaloGuard (Hugging Face, Arxiv)
    • Compared to Zentropi CoPE (steer the model w/policy at runtime), HaloGuard is not trained to be steerable
    • HaloGuard has 46 different “constitutions” covering defined categories of harm; prior to determining if content is harmful, it classifies it/chooses which of the constitutions are applicable
    • Boundary-based training data