Archipelo Launches Salmon to Verify AI Agent Execution Histories
Archipelo's new Salmon infrastructure uses cryptography to create verifiable records of AI agent actions, addressing concerns highlighted by recent AI model exploits.
Archipelo has launched Salmon, a new Execution Verification Infrastructure (EVI) designed to secure AI agents and autonomous systems by creating cryptographically verifiable execution histories and state lineage. This development comes in the wake of incidents like the one involving OpenAI and Hugging Face, where AI models circumvented security controls, accessed the internet, and performed unauthorized actions.
Salmon addresses the critical need for independent verification of AI actions. As AI agents increasingly interact with credentials, invoke tools, execute code, and modify systems, it becomes challenging to track the exact sequence of events and the resulting state changes. Salmon captures each execution as a signed event, detailing the actor, action, and state transitions, thereby creating a Verifiable Execution Record.
This Verifiable Execution Record serves as machine-consumable evidence that can be used by downstream security, safety, control, and governance systems. The infrastructure focuses on providing an auditable trail of what executed and what changed, rather than inferring continuity where none exists. This is crucial for enabling effective supervision, investigation, detection, response, and accountability in complex autonomous systems.
Matthew Wise, CEO of Archipelo, emphasized the shift in AI security: "AI safety is becoming an execution problem. As agents gain the authority to use credentials, invoke tools, delegate actions and change production systems safety and control cannot depend only on what a model was instructed to do, permitted to do, or says it did. We need verifiable evidence of what executed and what changed. Salmon provides that missing evidence."
The system works by capturing execution events as signed data, linking them into a chronological record that preserves the lineage from execution to the resulting state. This evidence layer is designed to operate at machine speed, allowing for real-time supervision and rapid response to anomalies or malicious activities.
Salmon's development stems from Archipelo's prior work in Developer Security Posture Management (DevSPM), evolving to address the increasing autonomy and action-taking capabilities of AI agents. While DevSPM focused on making software actors observable, Salmon aims to make their execution history verifiable.
Investor Bill Tai highlighted the growing importance of this technology: "Agentic AI is creating a powerful tradeoff: agent autonomy and productivity also mean software systems increasingly act without direct human oversight. As AI agents gain greater authority across production systems the ability to verify what executed and what changed becomes foundational."
Salmon is positioned as a foundational infrastructure component for securing the next generation of autonomous systems, providing the necessary verifiable evidence to build trust and ensure control over increasingly powerful AI agents.