San Francisco, USA, September twenty fifth, 2026, CyberNewswire
Archipelo at the moment introduced Salmon, Execution Verification Infrastructure (EVI) for AI brokers and autonomous programs, powered by a cryptographic protocol designed to make execution historical past verifiable.
Salmon establishes verifiable execution historical past and state lineage throughout people, AI brokers, and automation.
The launch follows the OpenAI–Hugging Face incident, through which OpenAI reported that fashions taking part in cybersecurity evaluations circumvented isolation controls, gained web entry, exploited vulnerabilities, accessed third-party programs, collaborated and delegated work throughout brokers, and took actions their builders didn’t direct.
OpenAI known as the incident a “warning shot” and stated subtle AI programs require safeguards that function on the velocity of the brokers themselves.
AI brokers use credentials, invoke instruments, execute code, name APIs, delegate work to different brokers, modify infrastructure and alter manufacturing state.
As these programs function throughout a number of actors and programs, the ensuing state doesn’t by itself protect the execution historical past that produced it.
Salmon is designed to seize execution as signed occasions and file the state transitions they produce, offering machine-consumable execution proof for investigation, detection and response, remediation, supervision and accountability throughout AI safety, security, management and governance programs.
Archipelo is backed by Dell Applied sciences Capital and expertise buyers together with Zoom CEO Eric Yuan, Andy Bechtolsheim, Invoice Tai, David Weisburd, Hack VC, Sangha Capital and Nima Capital.
Its staff brings expertise spanning NASA, DoD, AWS, Google, Cisco, Meta, Harvard, MIT and Berkeley.
Execution Verification for AI Safety and Management
Id establishes who or what’s performing. Authorization determines what an actor might do. Runtime controls decide whether or not execution ought to proceed. Observability screens habits.
Execution Verification addresses a distinct query: What executed, and might that historical past be independently verified?
Salmon is designed to offer an execution proof layer for autonomous programs.
From Execution to a Verifiable Execution File
Salmon captures execution as signed occasions and data the state transitions they produce. Every occasion identifies the actor, motion, state earlier than, state after and cryptographic signature.
Salmon hyperlinks these occasions right into a Verifiable Execution File, preserving the lineage between execution and ensuing state.
Seize → Execution File → State Lineage → Verification
The file produces Machine-Consumable Execution Proof that downstream programs can confirm and use robotically.
Salmon establishes the execution file. Different programs motive from it.
When execution proof is unavailable, Salmon preserves the hole moderately than manufacturing continuity.
Execution Proof for AI Safety and Security
“You can’t management autonomous programs with out verifiable proof of their execution,” stated Matthew Smart, Creator & Protocol Architect of Salmon and CEO of Archipelo.
“AI security is changing into an execution downside. As brokers achieve the authority to make use of credentials, invoke instruments, delegate actions and alter manufacturing programs security and management can’t rely solely on what a mannequin was instructed to do, permitted to do, or says it did. We want verifiable proof of what executed and what modified. Salmon gives that lacking proof.”
“Highly effective AI shouldn’t be answerable for establishing the historical past of its personal execution. Seize the proof when execution occurs. Make it cryptographically verifiable. Let AI security, management, safety and governance programs motive from the file.”
Salmon is meant to offer structured execution proof for machine-speed supervision, investigation, detection and response, remediation and accountability throughout autonomous programs.
Autonomous Techniques and Execution Proof
Verifiable execution historical past gives AI security, management, safety and governance programs with proof they’ll use as autonomous programs execute and modify software program and infrastructure.
“Agentic AI is creating a robust tradeoff: agent autonomy and productiveness additionally imply software program programs more and more act with out direct human oversight,” stated Invoice Tai, investor in Archipelo.
“As AI brokers achieve larger authority throughout manufacturing programs the flexibility to confirm what executed and what modified turns into foundational. Salmon gives the execution proof wanted to strengthen agent security, management, and governance.”
From DevSPM to Salmon
Salmon grew out of Archipelo’s R&D work on Developer Safety Posture Administration and software program actor attribution.
As AI advanced from helping builders to executing actions throughout instruments, credentials and infrastructure, Archipelo expanded its work from software program actor attribution to execution historical past and state lineage.
DevSPM made software program actors observable. Salmon makes their execution historical past verifiable.
About Salmon
Salmon gives verifiable execution historical past for autonomous multi-actor software program programs.
Salmon is Execution Verification Infrastructure for autonomous programs powered by a cryptographic protocol establishing verifiable execution historical past and state lineage throughout people, AI brokers and automation.
Salmon is developed by Archipelo, Inc.
Customers can be taught extra: salmon.programs
Stan Kapusta
Archipelo, Inc
