Arch Systems brings full factory AI to FedRAMP High via Knox

For years, Arch Systems' work inside U.S. defense plants stopped at the edge. Its software could sit close to the machines. The private-cloud half of its architecture, the part that pulls a mixed-vendor floor together with manufacturing execution system (MES) and enterprise resource planning (ERP) records, stayed out of reach for customers bound by federal security rules.

That changed on November 9. In a release distributed through PR Newswire, Arch said its Data Foundation and full Intelligence Layer of AI applications and agents can now run in FedRAMP Class D environments through a partnership with Knox Systems, a managed federal cloud provider. Each customer gets a single-tenant Arch instance inside Knox's environment that it owns and controls. Manufacturers that would rather use their existing private cloud can deploy there instead. Arch says the capabilities are generally available now to qualified federal, defense and regulated manufacturers.

The route matters as much as the destination. Arch didn't earn a standalone federal authorization; it moved into a boundary Knox already holds. For the operations and IT leaders at a defense contractor who would sign for this, the decision is now partly about Arch's factory software and partly about how Knox divides the compliance work.

The cloud layer defense plants were missing

Arch's case runs like this. The AI used in commercial factories depends on a data foundation that combines machine data with MES and ERP context, and that combination is what produces predictive insights and AI-guided actions on the floor. Defense work requires the foundation to sit in infrastructure the manufacturer owns and controls. Until now, the release says, contractors faced two options: build and run that infrastructure themselves, or do without.

Both chief executives describe the same bind. Andrew Scheuermann, CEO of Arch Systems, said defense manufacturers spent years standing up and running infrastructure of their own just to put AI to work on the factory floor, and that they "no longer have to choose between security and progress." Irina Denisenko, Knox's CEO, framed it from the cloud side: manufacturers have been forced to trade the AI tools they need against the infrastructure controls required by their missions. Between them, the pitch is easy to map: Arch brings the factory intelligence, Knox brings the controls.

Arch has worked in defense plants for years, but only at the edge; the Knox deal opens the layer that ties machine data to MES and ERP records, and lets controlled data flow into it. That job falls to the Connector Platform, which Arch says brings a floor of equipment from mixed vendors, together with its MES and ERP context, into one unified data model. Arch built its reputation in mission-critical discrete manufacturing, low-volume and high-mix production, and complex product types, and this is the first time defense customers can use its full hybrid architecture.

Arch pitches the defense offering as the same platform, applications and agents that run in what it calls the world's most advanced electronics and AI hardware factories. If that holds, a defense plant would be adopting the commercial product, not a separate federal edition with its own feature lag.

The bigger change is Controlled Unclassified Information. Arch says Knox shortened its authorization timeline and made it possible for Arch to handle CUI, so controlled data can enter the same unified model the AI agents run on. Edge-only deployments could analyze machine signals. A cloud data layer that accepts CUI could connect those signals to the records that make them useful, such as work orders and inventory, in a plant where much of that data is controlled.

The scope of that obligation is broad. In the Revision 2 text, NIST applied its requirements to every component of a nonfederal system that processes, stores or transmits CUI, or that protects those components. A cloud data layer receiving controlled work orders would be one of those components, which is why where it lives, and who secures it, becomes a compliance question and not just an IT preference.

How Knox's boundary carries the compliance load

Knox's FedRAMP Marketplace listing lays out the model. Application providers place their software inside the boundary Knox has authorized at FedRAMP High, and the listing says they inherit a substantial portion of the security controls they would otherwise have to meet. Knox's compliance team and its KnoxAI platform guide each provider from readiness through assessment to go-live. Once the application is live, Knox takes over ongoing compliance and the security of the boundary.

That process runs on KnoxAI. The listing says the platform measures a provider's security posture and control implementation against NIST 800-53 and other frameworks, points out gaps along with ways to fix them, and generates the documentation FedRAMP calls for. After an application goes live, the tool is supposed to keep the provider in compliance as its environment shifts, flagging emerging risks early so fixes stay within FedRAMP deadlines.

Isolation is built in. According to the listing, each application runs in its own dedicated, single-tenant account with no shared compute, data plane or storage. Workloads run only in U.S. regions of AWS, Azure or Google Cloud, with FIPS-validated encryption in transit and at rest and zero-trust access with full audit logging. Knox reports monthly to authorizing agencies, tracks remediation plans, and runs annual independent assessments and penetration tests. Every hosted application is individually assessed, listed on the marketplace, and under continuous monitoring from day one.

Knox also handles daily security operations. For each hosted application, the listing commits to round-the-clock coverage from both a security operations center and a network operations center. It also lists managed detection and response, ongoing scanning for vulnerabilities and configuration issues, and incident response, and Knox runs contingency and incident response exercises with each customer. For a plant IT team that has never staffed a federal-grade operations center, that list covers most of what renting the boundary really pays for.

Knox's own marketing makes the scarcity case. One of its resource pages notes that just 97 of the 547 certified cloud services on the FedRAMP Marketplace hold High authorization, and that Knox gets vendors to FedRAMP High in roughly 90 days. The same page cites federal cloud spending of $17 billion in 2024, with high-impact systems accounting for roughly 40% of it. Those figures come from Knox itself. Still, they explain why a factory analytics company would rent a boundary instead of building one: inheriting most of the controls cuts the work Arch has to own down to its application.

It helps to separate two kinds of evidence. The marketplace fields, including Knox's Class D (High) certification, its certification date and its authorization count, are the program's own record, and they also show the package as Rev5 on the Agency path. The service description beneath them is vendor-submitted text, which the marketplace labels as content FedRAMP does not review or endorse, so the inheritance and timeline claims rest on Knox's word.

Arch is also not an unusual tenant. The listing names the other offerings inside the same boundary, including Sift for hardware telemetry used by defense and aerospace contractors, Tractian for asset health monitoring, and Altana's supply chain network. Industrial data tools are clearly part of the mix. Knox's release boilerplate adds Adobe, Celonis, OutSystems, Armis and BigID to its roster, and PR Newswire also carries Knox announcements for Wraithwatch, an AI cyber defense company, at Class D and Craft, a supplier intelligence platform, at Class C.

One detail rewards a close read. The listing's single-tenant promise applies to each application's account, and some offerings in the boundary, including Sierra's conversational AI agents and Swimlane's Turbine, describe themselves as multi-tenant SaaS. Arch's release goes a step further, with a separate instance for every customer. Buyers comparing tools inside the same boundary should check which of those two models they are getting.

Owning the instance, sharing the work

The release says Arch becomes part of the customer's infrastructure, not a replacement for it, and that the customer owns and controls its instance. The listing also says Knox is directly accountable for the boundary's compliance posture. Both statements can be true together. In practice, though, responsibility is split three ways, between the manufacturer, Arch and Knox, and Knox says its Shared Responsibility Matrix documents who owns what.

That matrix, as described, splits duties among Knox, the application provider and the agency. A defense manufacturer is not an agency, so it should have its own column worked out before CUI goes in. Change control is the second question to settle. Arch promises the latest AI models and a growing set of agents, while the boundary is under continuous monitoring with monthly agency reporting, so a buyer should know how new features get approved inside it.

The listing suggests where that conversation starts. KnoxAI is pitched as the tool that keeps a provider compliant as its environment changes, and a new AI agent is exactly that kind of change. Asking Arch and Knox to walk through how a single new agent would move from build to approval would show how the process works in practice.

The best argument against reading too much into the Knox route comes from the release. Contractors that already run a private cloud they trust can deploy Arch there and get the same models, architecture and support. For them, the news is narrower: Arch's full hybrid platform is now offered on their own infrastructure, and the Knox authorization doesn't apply.

Impact levels are the next marker

FedRAMP High is a civilian-agency baseline. Knox's own guidance treats Department of War Impact Level alignment, formerly DoD, as a separate factor in choosing a platform. On that page, Knox lists Impact Level 4 as supported and Impact Level 5 as in process, with completion estimated for December 2026, and its site now promotes an IL5 landing zone.

The banner pitches that landing zone as a way to meet the Department of War's most critical missions in 180 days, twice the 90 days Knox quotes for FedRAMP High. Knox's comparison page also frames its IL4 support as a defense-aligned path on top of civilian agency coverage.

The same comparison shows the trade-off at the higher tiers. Knox lists Second Front Systems' Game Warden as IL5 authorized through the Defense Information Systems Agency, with coverage from IL2 through IL6, but requiring applications to be containerized on its Kubernetes-based infrastructure. SMX's managed services, by Knox's account, cover IL2, IL4 and IL5. Knox pitches its own model as deploy-as-is with no containerization mandate, which suggests its route trades a no-rework start for a defense impact level that is still in process.

Arch's announcement covers the FedRAMP Class D deployment and does not claim an Impact Level for its own instance. The contractors most likely to act first are those with CUI on the floor, mixed-vendor equipment and Arch already running at the edge. For programs that need a defense Impact Level, the next thing to watch is whether Knox's December IL5 milestone carries over to the factory AI tenants inside its boundary.