Most boardroom discussions about artificial intelligence center on upping the ante on efficiency, speed and competitive advantage. While such discussions should happen, the more urgent conversation needs to be around AI dependency in critical systems, suppliers, data flow, models and third parties. 

Executive should be thinking about increasing AI risk as the technology indirectly arrives through cloud platforms, software as a service, managed service providers, data processors and vendor tools that now include AI features by default. 

The inability to take stock of such passive AI integration into workflows can spiral into governance issues: 72% of executives say AI has been scaled across most or all business initiatives, while only around a third say the right controls are in place.

Two continuing weaknesses of supply chains are fragility and inherited supplier risk, which can be exploited by threat actors within a supplier’s environment.

Supplier-driven AI ecosystems have a similar threat profile. A single AI-enabled workflow has diverse building blocks, including cloud infrastructure, application providers, foundational models, application programming interfaces and subcontractors.

A weakness exploited in just one can quickly spread to others, poisoning the workflow.

AI becomes a supply chain issue when the board focuses solely on the use case rather than the dependency chain that drives return on investment. There’s no plan in place if the model changes without notice, a provider becomes unavailable, a vendor uses data in a way that violates an organization’s regulatory framework, or a regulator asks how a particular AI-influenced decision was made.

Looking at AI through the wrong lens can make you lose sight of supply chain exposure. The review process might go through the vendor contract with a fine-tooth comb but completely miss the model-change process. When it goes through legal, the team will examine data terms but miss operational dependencies. The security team will focus on monitoring systems and not AI prompts or outputs. The risk framework will drill down on internal AI use but will not examine how suppliers use AI on the organization’s behalf.

Over time, these gaps widen and continue to build exposure. As AI becomes more embedded in the decision-making process and across third-party relationships, organizations will be expected to show that they not only use AI responsibly, but are well aware of who controls the data it touches and how AI-influenced decision-making can be explained in detail.

Traditional supply chain disciplines should be incorporated into the AI ecosystem. Govern it as you would a critical supplier, operations system, or regulated process.

Best practices for governing a connected AI system include the following:

Secure complete visibility into the AI dependency chain. A clear picture of direct AI usage and indirect use through suppliers, platforms and partners is necessary. Don’t forget the movement of data, including what’s shared, the providers involved, the decisions it influences, and the various processes that depend heavily on AI-enabled output.

Ensure accountability. Every AI use case should be assigned to an owner responsible for human oversight, monitoring, escalation if things go wrong, and retirement of AI workflows. If the board needs answers for what the AI system does, data usage, dependencies, or problems, this owner should have the answers.

Manage third-party risk. Put AI providers and AI-enabled suppliers through the same risk management framework applied to critical vendors. This means asking practical questions: Are we becoming too dependent on one provider? Who are their subcontractors? How will our data be used? Can we audit what matters? Will we be told when the model changes? What happens if the service fails? And can we exit without disrupting the business?

Engage in impact-based governance. All exposures are not equal. Each AI use case must be scrutinized accordingly. A productivity tool that summarizes meeting notes doesn’t have the same exposure as one that affects supply chain operations.  The higher the business impact, the stronger the governance framework should be.

Strengthen information controls before deployment. Before AI is introduced into workflows, it’s imperative that organizations clearly define which data could enter external AI systems, the conditions that enable such data movement, and the protections governing it. Unclear vendor terms or informal experimentation shouldn’t put data at risk.

Focus on resilience and performance. There’s a tendency to judge AI use cases with performance metrics. But resilience should also be factored in. If providers alter AI models, restrict access, face regulatory action or change their terms of engagement, fallback options must be in place. This can take the form of alternative suppliers or manual overrides.

The clear and present risk with AI is that the organization will become heavily dependent on systems and suppliers into which it has no clear visibility, with no plan B if something goes astray. 

The focus, therefore, should be on seeing AI governance through the prism of supply chain governance. Boards that adopt this stance will scale AI adoption safely, baking in accountability, assurance and resilience before dependencies creep in.

Steve Durbin is chief executive of the Information Security Forum.