Sergio

😀Excited to share our new paper: Quantifying Trust: Financial Risk Management for Trustworthy AI Agents. arxiv.org/pdf/2604.03976 featured in @FortuneMagazine: fortune.com/2026/04/08/agent… Together with @Chi_Wang_, @tianyi_peng , @virtuals_io, and @t54ai, we propose the Agentic Risk Standard (ARS), a settlement-layer protocol that uses escrow, underwriting, and collateral to provide enforceable guarantees for transactions involving AI agents I’ve been working on agent infra since 2024, and most current works on agent infra are about Docker environments or efficiency. But is that all we need for infra for AI agents? Agents are already participating in real economic activity: providing services, making purchases, moving money. That means agent infra is not just technical, it’s also financial. Right now, using agents is basically “risk-on by default”: If things go well → everyone benefits If things go wrong → the user eats the loss That asymmetry is the real problem. 👀👀So I started thinking: what would finance for AI agents look like? We propose 👉Agentic Risk Standard (ARS)👈: 1⃣ Quantify agent risk 2⃣ Introduce escrow, premiums, collateral, underwriting, claims 3⃣ Shift from “try it and see” → “clear accountability and compensation” ⛑️Trustworthy AI shouldn’t stop at “models are reliable enough” But rather, it should answer: Who pays when things fail? How much? Under what rules? ARS moves us from model-level reliability to product-level accountability, from implicit trust to enforceable guarantees. ARS is not AI safety (which tries to make models behave) and not fintech with an AI skin (which keeps the human as the backstop). It's a risk layer: escrow holds payment, underwriting prices risk, collateral keeps agents accountable when things break. 🤗 The Agentic Risk Standard is an open-source initiative. We invite researchers, developers, and risk managers to contribute to the protocol and expand the agentic economy. paper: arxiv.org/pdf/2604.03976 github: github.com/t54-labs/AgenticR… website: t54.ai/ars