Urgent.News

What's breaking now, across thousands of outlets.

AI

SAAX Protocol: A Commitment Layer for Agentic Commerce

When an AI agent buys something and the purchase goes wrong, the parties usually discover the same problem: there is no structured record of what was actually committed. The authorization was valid. The credentials were real. The merchant fulfilled a legitimate order. The buyer is looking at a charge for something they never intended to purchase. Card networks were built for a two-party dispute.…

When an AI agent makes a purchase, it can be difficult to determine whether the agent actually committed to the purchase and if the merchant fulfilled it. This is due to the lack of a structured record of commitments in the current authorization and settlement layers. The SAAX Protocol aims to address this issue by creating a vendor-neutral commitment lifecycle for autonomous commerce.

The SAAX commitment records five things before execution begins: authority, terms, acceptance criteria, evidence requirements, and recovery policy. These records are issued before payment and evidence is attached as references. The protocol also defines outcome and finality states for each commitment.

A unique aspect of the SAAX Protocol is its handling of failed agent attempts. If the agent's first attempt leaves the machine without a response, the protocol rule is that an unknown outcome is a status check, not another POST request. The agent queries the commitment state before retrying, with possible outcomes including safe to retry, waiting, fulfilled, or failed. On the merchant side, they must write in_flight to durable storage before processing the request to prevent double processing.

The reference implementation of the protocol includes a Spec, conformance suite, documentation, and a MCP endpoint for testing. The protocol is vendor-neutral and open for comment, making it suitable for agents that transact across various rails and systems. The Router is one implementation of the protocol, using x402 and MPP for settlement, but the Protocol itself does not require any specific rail.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Build an AI-powered extension that helps students learn more effectively

Build an AI-powered extension that helps students learn more effectively. Features to add AI Tutor — answers questions based on course content.

  • AI-powered extension enhances student learning experience
  • Features include AI Tutor, Smart Quiz Generator, and voice learning
  • Open-source models offer transparency and customization

TouchGrass: An AI Whose Success Is Measured by How Quickly You Stop Using It 🌱

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass What I Built TouchGrass is an AI-powered application built for the Hacktoberfest ’26 Week 1 challenge, Touch…

  • TouchGrass AI encourages outdoor activities over device usage
  • Users customize mission details like activity, duration, and difficulty
  • Open-weight model ensures adaptable mission generation

Route Once, Fail Over Among Equals: When NOT to Retry an LLM Call

Most LLM APIs I review have one IChatClient and a prayer. Hello goes to the same model as "why does this async code deadlock under load." When that provider returns 503 for twenty minutes, so does the…

  • Only fail over among models of equal quality or reliability.
  • Do not retry when provider returns 400, indicating malformed request.
  • Commit only once when streaming the first token to avoid half-answer.

Around 1.000 AI products don't appear in corporate access systems

A Reco analysis based on the 2026 State of Agent Security Report examines how AI assistants enter company environments through third-party products.

  • Around 1,000 AI-powered digital assistants may be undetected in company systems
  • Only 282 of 1,280 studied AI products are protected by single sign-on systems
  • Companies urged to verify assistants, limit permissions, and clarify their actions

I built an open-source, self-hosted take on ElevenLabs' Reception.ai

The phone rings at a dental clinic while everyone's hands are busy. Nobody picks up. The caller books somewhere else. Frontdesk.ai is an open-source AI receptionist that picks up instead.

  • Frontdesk.ai is an open-source AI receptionist for phone calls.
  • Inspired by Reception.ai, it's an independent project not affiliated with ElevenLabs.
  • Runs on user's system via Docker Compose with Next.js, Supabase, LiveKit, and more.

More from Saturday 10 October →