VIAS
The evidence plane for autonomous agents
VIAS receives and seals submitted execution events outside the agent runtime, creating tamper-evident evidence that can be verified independently of the platform that acted.
Control planes decide what agents may do. VIAS preserves the record of what they did.
What VIAS Is
VIAS is an execution-evidence layer that operates separately from the agent runtime and decision environment. At the platform boundary, the operator or its observability stack submits execution events to VIAS, where they are evaluated against the registered authority grant and sealed into a tamper-evident evidence chain. Where an outcome falls outside the grant, the original record remains unchanged and any correction, reversal or reconciliation is cryptographically linked as a new event.
VIAS combines established cryptographic, policy and evidence-management primitives at the point where accountability must live: between autonomous action and the durable evidentiary record. Because the evidence function is separated from the model, planner, runtime, orchestration system and tool caller, auditors and counterparties do not have to rely exclusively on records produced by the system that made the decision. Evidence is sealed separately from the system that acted. The owner may be the same.
How It Works
A submission-to-verification evidence system.
Agent & Tool Boundary
VIAS Governance Boundary
01
SUBMIT
Submitted and received outside the agent runtime
02
ATTEST
Rules-based comparison against the registered authority grant
03
SEAL
Append-only evidence chain commitment
Auditor / Regulator / Counterparty
01
SUBMIT — Platform Boundary Routes
The platform boundary routes a normalised execution event to VIAS. The evidence package identifies the relevant authority, requested action, execution payload, executor assertion and reported result. Deployments claiming comprehensive coverage must ensure that applicable actions cannot bypass the configured evidence boundary.
02
ATTEST — Governance Gate Comparison
The VIAS Governance Gate applies a rules-based comparison against the registered authority grant — not a judgement of whether the action was right. The evidence record distinguishes between facts VIAS records itself (such as the time it received a record), assertions supplied by upstream systems, policy comparisons derived by VIAS, and external attestations supplied by trusted issuers. This distinction preserves the chain of responsibility — VIAS protects the integrity of the record, while the originator of each upstream assertion remains responsible for its reliability.
03
SEAL — Ledger Committed
VIAS cryptographically seals the record into an append-only evidence chain. The original event is never silently rewritten — corrections, reversals and reconciliations are recorded as separately sealed events linked to the original evidence. Where the comparison shows the action fell outside the registered grant, VIAS records the discrepancy and any governed response in the same chain. VIAS does not silently replace the original event or imply that the underlying external effect has necessarily been reversed.
Verification scope: A valid seal establishes that the sealed record has not subsequently changed. It does not, by itself, prove that every upstream assertion was factually correct or that every applicable action was submitted.
04
EXPORT — Evidence Pack
Sealed records and their coverage statement are exported as a self-contained evidence pack.
05
VERIFY — Offline
A reviewer is designed to check signature, sequence and vectors with the published public key, offline, without a live VIAS service. The signed-export verification chain is pending end-to-end testing.
Verification scope: Offline verification is designed to work from the exported pack alone. Until the signed-export chain completes end-to-end testing, verification should be treated as designed-for, not confirmed.
Every record checked against all six.
- Input
- What the agent was given to act on
- Origin
- Which model and version acted, and where the record came from
- Response
- What the agent produced
- Handoff
- How the action links to other agents in a chain
- Scope
- What the agent was authorised to access, and what access was reported
- Authorisation
- Who approved the action, and under which policy
Out-of-band from the agent is not the same as independent of the vendor
Hardware containment tools from the leading chip vendor are, as we read their public description, control-plane measures — valuable and increasingly required. But they run on the vendor's own silicon and are verified by the vendor's own tooling. They are out-of-band from the agent, but not independent of the vendor.
VIAS evidence is sealed independently of both the agent runtime and the underlying compute vendor. The custody boundary is the actor, not the infrastructure provider.
What It's Built For
As autonomous agents begin making purchases, changing records, operating enterprise systems and exercising delegated authority, the evidence question becomes an operational and liability question. Organisations need six questions answered before they can stand behind an agent's action:
- 01What authority existed
- 02What action was requested
- 03What payload was executed
- 04Which executor was involved
- 05What result was reported
- 06Whether the evidenced action remained within the registered authority grant
VIAS provides evidentiary custody for those questions without inspecting private reasoning or asking one model to judge another.
EU AI Act
The EU AI Act applies progressively, and its timelines may change under the EU’s omnibus simplification package. VIAS sealed records are designed to support technical logging and record-keeping under Article 12, post-market monitoring under Article 26(6), and incident investigation under Article 73. Applicability depends on the system’s classification and the organisation’s role.
ISO 42001
ISO/IEC 42001:2023 specifies requirements for establishing, implementing, maintaining and continually improving an AI management system. It addresses responsible AI use, risk management, traceability, transparency and reliability. VIAS is designed to provide evidence that can support an organisation’s wider AI management system.
NIST AI RMF
The NIST AI Risk Management Framework is a voluntary, non-sector-specific framework designed to help organisations manage risks associated with AI and incorporate trustworthiness into the design, development, use and evaluation of AI systems. VIAS evidence can support governance, measurement and accountability practices within an organisation’s implementation of the framework.
UK ICO
The ICO Tech Futures: Agentic AI report examines emerging agentic-AI capabilities and related data-protection questions, including accountability, complex automated decision-making, transparency and cybersecurity. The ICO explicitly describes the report as early thinking rather than formal guidance or regulatory expectations. VIAS is designed to support accountable evidence capture where agentic systems process information and perform consequential actions.
Important qualification: Specific obligations, implementation dates and applicability depend upon the organisation, autonomous system, deployment role and jurisdiction. VIAS provides evidence infrastructure, not automatic legal compliance.
VIAS connects at the governance layer, designed to integrate alongside major agent platforms, with no re-architecture of the agent platform required. Evidentiary independence reduces platform lock-in and avoids dependence on a specific model or agent runtime. Platform-side configuration is needed to route execution records to VIAS.