In this article, learn about:
What the Pentagon's Anthropic ban means for supply chain risk
The limitations of AI as a data replacement tool
How supply chain risk management can adapt to an AI era
In March 2026, the Pentagon issued a directive ordering military components to remove all Anthropic products from their systems within 180 days, citing the AI company as a supply chain risk under national security authority. For defense contractors, the challenge was both technical and legal, and immediate.
For enterprise retailers, the lesson should land just as hard, even without a government mandate in sight.
The Anthropic ban is a signal, not an anomaly. Governments, regulators, and major enterprise customers are starting to treat AI tools the same way they treat any other embedded vendor: as components that carry risk, require governance, and can be restricted or removed without much warning.
That raises an uncomfortable question for CIOs and CISOs at enterprise retailers: if your organization were required to identify and remove a specific AI tool tomorrow, could you do it?
Most organizations today cannot. The gap between how broadly AI has been adopted and how poorly it is governed is rapidly becoming its own category of supply chain risk.
The Pentagon's AI Ban as a Supply Chain Problem
The Pentagon's directive wasn't just a vendor dispute. By classifying Anthropic under supply chain risk authority, the administration put a name to something security professionals have been warning about for years: AI tools are not neutral software utilities. They are embedded dependencies, woven into workflows, applications, and systems in ways that are often invisible to the organizations using them.
How AI Enters Your Environment Untracked
As CSO Online reported, AI models can enter an organization's environment through multiple channels, most of them untracked by central security teams: directly through APIs (application programming interfaces, which allow software systems to communicate with one another), embedded inside internally developed applications, introduced through developer tooling and code libraries, or bundled inside third-party software and vendor-provided systems.
The same week an organization rolls out a formal AI governance policy, a product team may have already integrated a model into an inventory management system three layers deep.
The Numbers Behind the Visibility Gap
Cisco's 2025 AI Readiness Index found that only 27% of organizations report having granular access controls over AI systems and datasets. Nearly three-quarters of enterprises cannot clearly see which AI tools are operating in their environment or what data those tools are accessing.
For retailers managing complex supplier networks, that is a structural vulnerability hiding in plain sight.
Shadow AI in Retail Operations
Shadow AI refers to AI tools being used within an organization without the knowledge or approval of the IT or security team. It is the AI equivalent of shadow IT, and it carries similar risks at a larger scale.
In retail, shadow AI adoption tends to happen function by function, without coordination. Merchandising teams use AI to model demand forecasting and assortment planning. Logistics teams use it to optimize routing and carrier selection. Procurement teams use it to evaluate trading partners and review purchase orders. Finance teams use it to flag invoice discrepancies and chargebacks. In most cases, each initiative launches independently, with no shared inventory of what tools are in use or what data is flowing through them.
Governance is Not Keeping Pace
IDC forecasts that by 2026, most CIOs will need to diversify security strategies specifically to address supply chain and generative AI risks. That diversification can only happen with visibility, and visibility requires a deliberate, centralized approach to cataloging AI dependencies before a ban, a breach, or a compliance failure forces the issue.
The 2026 Deloitte State of AI in the Enterprise report adds another layer: only one in five companies has a mature governance model for autonomous AI agents. Agentic AI, systems that take actions and make decisions with minimal human intervention, is already deployed across supply chain functions. Without governance, those agents are making operational decisions using data and logic that no one in the organization has reviewed or approved.
Why AI Is Only as Good as the Data Feeding It
Every AI model reflects the data it was trained on, and every output it produces reflects the data being fed into it in real time. That is what makes AI governance a supply chain problem, not just an IT problem.
Clean Data as a Prerequisite
An AI model used for production scheduling or distribution routing is only accurate if the underlying data is clean, current, and normalized. Data normalization, the process of structuring data from multiple sources into a consistent format so it can be compared and analyzed, is a prerequisite for any AI system operating across a multi-supplier environment. Without it, AI outputs reflect the chaos of inconsistent data rather than the logic of a well-tuned model.
The Supply Chain Management Review's 2026 analysis confirmed this pattern: in 2025, AI delivered measurable value where data was clean, governance was clear, and risk was managed deliberately. Where data foundations were weak, AI amplified the problem.
The Ceiling of Internal Data
A model trained on your own historical data can identify patterns within your operations. A model connected to a broader network of anonymized, benchmarked industry data can tell you what good looks like across hundreds of thousands of trading partner relationships, across different retailer types, across different categories and seasons. These are fundamentally different capabilities, and the difference between them is the data.
The Build vs. Buy Question for Supply Chain AI
The debate over whether to build proprietary AI solutions or purchase commercial ones is not new, but AI has made building faster and cheaper, which has led more technology organizations to revisit the question.
For supply chain specifically, the answer depends on where your organization actually competes.
When Building Makes Sense
Some retailers have built a proprietary capability, a unique fulfillment model, a pricing engine, a private-label supply chain, that is itself a source of competitive advantage. For them, building core AI systems can make sense. The edge is in the system, and the investment to build and maintain it is justified by what it creates.
When Integration Is the Advantage
Most enterprise retailers are not in that position. They compete by connecting and streamlining processes across a complex value chain of third-party logistics providers, suppliers, and retail partners more effectively than anyone else. Their advantage is not in any single proprietary system. It is in how well all the systems work together.
For that second group, the build option has a ceiling that AI cannot raise: your model only knows what your organization has seen. AI can make it faster and cheaper to build a demand signal model or an inventory optimization tool. What it cannot do is tell you what "good" looks like.
Knowing that your order fill rate is 94% is only useful if you know whether 94% is competitive, lagging, or leading for your category and your trading partner mix. That context does not live in your transaction history. It exists in two decades of trial, error, and refinement across hundreds of thousands of supplier relationships, and in the ability to see, in real time, how comparable businesses are actually performing.
A retail network that connects tens of thousands of suppliers, retailers, and logistics providers generates that picture continuously: where the bullwhip effect is creating inventory distortion, what on-time, in-full delivery rates look like across comparable supplier relationships, and where the gaps are. No proprietary model can replicate that.
The Hidden Risk of Building Your Own
A proprietary AI system built in-house is itself a vendor you now manage, maintain, and are solely responsible for when it fails or when the data it was trained on becomes stale. That is a different kind of supplier risk, and most CISOs have not yet added it to their list.
What a Supply Chain Risk Management Plan Should Include for AI
Supply chain risk management programs have traditionally focused on physical disruptions: natural disasters, port congestion, supplier financial instability, geopolitical events. AI governance needs to be part of that framework now, not eventually.
What Industry Guidance Recommends
Corporate compliance guidance for 2026 recommends three immediate actions: creating an internal registry of all AI use cases across the organization, reviewing liability and audit rights in existing contracts with AI providers, and updating standard operating procedures for data handling wherever AI is involved.
The NSA's March 2026 guidance goes further, framing AI as a layered supply chain where weaknesses at any layer can disrupt operations. Their recommendations include requiring cryptographic signing across model lifecycles, mandating adversarial testing, and building recurring audits into AI vendor contracts.
Five Areas Every AI Risk Framework Should Cover
Inventory and visibility
Catalog every AI tool in use across the organization, including those embedded in third-party software and vendor-provided systems. A software bill of materials (SBOM), a structured inventory of the components that make up a software system, is the starting point. An SBOM that does not account for AI components is incomplete.
Data governance
Establish clear policies for what data can flow into AI systems, especially those operated by external vendors. Supplier records, inventory data, and customer data carry different risk profiles. Know which AI tools have access to each.
Supplier risk evaluation
Treat AI vendors with the same scrutiny applied to other strategic suppliers. Review their financial stability, compliance posture, and data security practices, and plan for what would happen to your operations if they were suddenly removed.
Business continuity planning
A continuity plan that does not account for AI disruption has a gap. If a critical AI tool is banned, breached, or discontinued: What is the fallback? How long would operations be affected? Who owns that decision?
Continuous monitoring
Static assessments are not enough. Supply and Demand Chain Executive notes that risk frameworks must adapt as internal and supplier AI usage changes, regulations evolve, and new data dependencies emerge.
Why a Retail Network Provides What AI Alone Cannot
AI models, no matter how sophisticated, operate within the boundaries of the data they have access to. For a retailer trying to understand whether its fill rates are competitive, whether its lead times reflect current market conditions, or where its suppliers are underperforming relative to peers, the model is only useful if it has context beyond the organization's own history.
An AI system trained on one retailer's transaction history can find patterns in that data. It cannot tell you whether those patterns represent a problem or an industry-wide norm. Performance benchmarks are only meaningful when grounded in real operational data from across the market.
SPS Commerce connects more than 120,000 companies across retail, grocery, distribution, supply, and logistics. That network reflects real transactions across trading partner relationships of every type and scale, two decades of operational refinement across supplier compliance, EDI (electronic data interchange, the standardized transmission of business documents between systems), and fulfillment, and live performance benchmarks that reflect what the market is actually doing.
Accessing that intelligence through SPS Commerce Fulfillment gives retailers visibility into what operational excellence looks like in practice. A prediction tells you where you might be. A benchmark tells you where you stand. That is why a connected retail network is a more durable foundation for supply chain decision-making than any AI tool built in isolation.
