.png&w=828&q=75)
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.
10 May 2026

HayStack turns online writing into a paid knowledge market for AI agents. Writers can publish or import articles from RSS, configure per-post access policies, and earn whenever an AI agent reads the full article. Humans can still discover and read open or AI-metered content, while agents encounter an x402-style payment flow that requires payment before full access. The platform has two main surfaces. HayStack Publisher gives authors a Substack-like editor, RSS import flow, per-post paywall controls, and a dashboard showing human revenue, AI read revenue, and live settlement events. HayStack Agent is a Gemini-powered research agent that searches article previews for free, evaluates relevance against a budget, pays for selected full reads, and synthesizes an answer with citations. For payments, HayStack uses Circle Developer-Controlled Wallets on Arc testnet. A funded agent treasury wallet pays the writer’s Arc testnet wallet per read in USDC. The backend stores Circle provider transaction IDs, settlement status, and Arc transaction hashes, then streams settlement events to the frontend via SSE. The article API implements x402-shaped behavior: unpaid agent reads receive HTTP 402 with preview and payment metadata, while successful payments return a signed X-Payment receipt that unlocks full content. This is not a RAG demo over a static corpus. HayStack is a live economic loop for agentic content access: search is free, full reads are paid, writers control pricing, and value moves through real sponsor rails.
26 Apr 2026