Salmon launches infrastructure to verify AI agents' execution history

Salmon introduced Execution Verification Infrastructure, a new category aimed at making AI agents' consequential actions cryptographically verifiable across software systems. The pitch matters because autonomous systems are moving from generating outputs to changing state, creating a need for evidence of what executed, under whose authority and what changed.

Why it matters: - AI agents are moving from producing outputs to taking actions inside production software, infrastructure and credentials. - That shift creates a new need for cryptographic evidence of what executed, who or what had authority, and how system state changed. - Salmon is aimed at giving security, audit, governance and engineering teams verifiable execution history instead of after-the-fact reconstruction.

What happened: - Salmon introduced Execution Verification Infrastructure, or EVI, on Sept. 24, 2026. - The infrastructure category is designed for AI agents and autonomous systems that act across tools, applications and infrastructure. - Salmon is developed by Archipelo. - Archipelo says its protocol creates verifiable execution history across software systems. - Archipelo also said the company is backed by Dell Technologies Capital and investors including Zoom CEO Eric Yuan, Andy Bechtolsheim, Bill Tai, David Weisburd, Hack VC and Nima Capital.

The details: - The Salmon protocol captures supported execution through signed execution events. - It records state transitions with actor and authority context. - It preserves execution history and state lineage. - It supports cryptographic verification and replay. - Salmon describes the result as a Verifiable Execution Record. - The company says downstream systems can use Machine-Consumable Execution Evidence for AI, security, engineering, audit and governance workflows. - Salmon does not replace existing AI safety, control, cybersecurity or governance systems. - The company positions EVI as an underlying layer those systems can consume. - Archipelo says the problem is not just what an AI system was authorized to do, but what it actually executed and what changed as a result. - The company says reconstruction from model outputs, telemetry, logs or the actor’s own account may not be enough after the fact. - Salmon is designed to capture evidence as execution happens. - The product is intended to make that evidence available for later verification and replay. - Archipelo said its team has experience spanning NASA, the U.S. Department of Defense, AWS, Google, Cisco, Facebook, Harvard and MIT. - The company says it has raised $12 million. - Archipelo also lists Sangha Capital among its backers.

Between the lines: - The launch reflects a broader shift in AI infrastructure from monitoring behavior to proving action. - That distinction matters because observability tools can show what appears to have happened, while Salmon is trying to preserve evidence of what did happen. - The company is framing execution verification as a foundation layer for autonomous software, not a replacement for existing controls. - Bill Tai, an investor in Archipelo, said agentic AI creates a tradeoff between autonomy and oversight, and that the ability to verify what executed and what changed becomes foundational. - Archipelo traces Salmon to its earlier work on Developer Security Posture Management and software-actor attribution. - As AI systems moved from assisting developers to executing consequential actions, the company says the problem expanded from observing software actors to verifying execution. - The company argues that even well-behaved autonomous systems create attribution, debugging, audit and accountability needs once they can change software state. - Matthew Wise, creator and protocol architect of Salmon and CEO of Archipelo, said execution evidence should be captured at execution time and made cryptographically verifiable rather than reconstructed afterward.

What's next: - Archipelo is positioning Salmon for frontier AI, cybersecurity, cloud and developer infrastructure, enterprise AI and high-consequence autonomous systems. - The company says machine-speed systems need machine-consumable execution evidence. - Salmon is being developed to make cryptographically verifiable execution history and state lineage a foundational property of autonomous software systems. - Archipelo’s LinkedIn page is available here.

The bottom line: - Salmon is betting that as AI systems gain authority to act, proof of execution will become as important as policy, monitoring and authorization.