@emanpiece86244

GuardianRail is a representation-aware action firewall for open-weight customer-operations agents. The project shows a regulated support agent running on Gemma 3 12B IT with Gemma Scope 2 sparse-autoencoder features monitored during inference on AMD MI300X. Instead of only checking the final text response, GuardianRail displays the operation the agent is about to take, reads internal safety-relevant features, gates that proposed action, and logs the evidence behind every allow, block, or escalation. In the demo, a normal support request is allowed, a prompt-injection attempt is blocked before a restricted action can run, and a social-engineering request is escalated to human review. The Streamlit interface shows live safety signals, feature thresholds, the Action Firewall decision, policy-layer clamp/boost interventions, GPU usage, and a SQLite audit trail. The goal is not to claim jailbreaks are solved; it is to make open-weight agent safety observable, tunable, and auditable for teams deploying agents on their own infrastructure.

Grab your ideas, spin up compute, and start building what’s next! Start with $100 in AMD Developer Cloud credits and begin building right away. Get access to hands-on tutorials, workshops, and curated learning resources to help you build faster and go further. To unlock credits and join the hackathon, sign up for the AMD AI Developer Program. 🤝 Go solo or team up — ship something real. 💰 $21,500+ total prize pool & 1 × AMD Radeon™ AI PRO R9700 GPU up for grabs. 📍 On-site Venue (May 9-10): MindsDB SF AI Collective 3154 17th St, San Francisco, California, USA 📍On-site participation is by invitation only. Travel and accommodation expenses will not be covered. 🧑‍💻 Developers, builders, and innovators - bring your boldest ideas and let’s turn them into real AI systems on AMD.