Security, Architecture & Compliance

ESS has completed SOC 2 Type I audit fieldwork and passed its internal control checks. The auditor's report is expected September 2026 and will be available under NDA; until it is issued, ESS does not claim certification. SOC 2 Type II, which tests that those controls operated effectively over a period rather than at a point in time, follows as the observation window completes. A SOC 2 report describes the controls an organization actually operates — access management, change management, encryption, monitoring, incident response, vendor management and business continuity — as tested by an independent auditor, rather than as described by the vendor. The distinction that matters in a security review is Type I versus Type II: Type I attests that controls are suitably designed as of a point in time, while Type II tests that they operated effectively across an observation window. Type II cannot be rushed, because the window has to elapse, so the honest sequence is Type I first and Type II as it completes. Ask any technology partner which one they hold, then ask for the report itself and read the exceptions section; that is where a SOC 2 becomes informative rather than decorative. ESS supplies the report, the scope statement and the management assertion to prospective clients under NDA as part of a standard RFI response.

Inside infrastructure you control. ESS Fusion AI is deployed into your own Microsoft Azure or AWS tenant rather than onto a shared multi-tenant service, so ingestion, document data extraction, validation and record creation in your ERP all happen within your existing security boundary, your existing region and your existing network controls. That means data residency is whatever you have already decided it is — if your tenant is pinned to a region for regulatory reasons, so is the platform. For organizations whose mandate does not permit outbound connectivity at all, ESS deploys a fully air-gapped, self-hosted configuration with a local vision model and no external calls whatsoever.

No. Your documents, your extracted data and your corrections are never used to train public AI models, and they are never pooled with other customers' data. Model improvement inside your deployment is scoped to your deployment: when a reviewer confirms or corrects an extraction, that signal strengthens accuracy for you, in your tenant, and goes nowhere else. This is a structural property of deploying into your own infrastructure rather than a policy promise layered on top of a shared service — in an air-gapped install there is physically no path for data to leave. ESS supplies the model governance documentation, including which models are used and how versions are pinned and changed, during a security review.

Your process design controls this, not the model. Every document carries a confidence score per field, and anything below your configured threshold routes to a reviewer before it reaches the ERP rather than posting and being corrected afterwards. Extracted data is validated against live ERP master data first, so unknown customers, invalid part numbers and prices outside tolerance are caught before a record exists. Approval routing and segregation of duties follow the financial controls you already operate. Every document then retains a complete audit trail — the source file, what was extracted, the confidence each field carried, who reviewed it, what they changed and when — so both internal and external audit can reconstruct exactly how any given record entered the ERP.

A complete written response rather than a sales call. During an RFI, ESS supplies the deployment topology and data-flow diagram, the data retention and deletion model, the model governance documentation, the per-document audit trail design, the ERP integration and credential-handling approach, the subprocessor list for your chosen deployment model, business continuity and disaster recovery documentation, penetration testing and vulnerability management summaries, certificates of insurance, and the SOC 2 report once issued. Most of this is supplied under NDA because it should not be published on a marketing page. If your organization has its own questionnaire — a CAIQ, a SIG, or your own internal template — send it and ESS completes yours rather than returning a generic pack.

Yes. The air-gapped configuration runs a self-hosted vision model entirely inside your network with no external connectivity, which is the right answer for defense, critical infrastructure and any environment operating under a mandate that prohibits outbound calls. It removes most of a standard security questionnaire rather than answering it item by item, because the questions about third-party data transmission, subprocessors and cloud egress no longer apply. The trade-off is that you provide the compute and take on the update cycle for the local model; ESS sizes that during the assessment so it is a known cost rather than a surprise after go-live.