Manufacturers have moved quickly to embed AI into production planning, quality control, and supply chain execution. But new research by Kiteworks suggests that while these systems are closely watched for reliability, they remain exposed to more hostile and regulatory-driven risks that are likely to intensify in 2026.
New manufacturing-specific findings from Kiteworks’ Data Security and Compliance Risk: 2026 Forecast Report point to a widening gap between how industrial AI is operated and how it is governed. The study shows that manufacturers have invested heavily in oversight mechanisms designed to prevent accidental failures, yet have made far less progress in preparing for adversarial attacks, regulatory scrutiny, and cascading AI failures across supply networks.
Reliability First, Threat Readiness Later
According to the report, manufacturing organizations outperform most other sectors when it comes to operational AI controls. Roughly 63% maintain human oversight of AI systems, and 56% use AI data gateway monitoring to track performance in production-critical environments. These measures reflect long-standing industrial priorities around uptime, safety, and process stability.
Where manufacturers fall behind is in anticipating intentional abuse of AI models. Only 7% of respondents reported conducting AI red teaming or adversarial testing, less than half the global average. That leaves the vast majority of industrial AI systems untested against threats such as model poisoning, manipulated training data, or inference attacks designed to extract sensitive information.
“Manufacturing has built AI governance for reliability, not hostility,” said Tim Freestone, chief strategy officer at Kiteworks. “That works when failures are accidental. It fails when threats are intentional. AI systems don’t just break. They get attacked.”
Recent industry analysis shows that adversarial techniques are becoming more accessible as AI adoption scales, increasing the likelihood that attackers will target operational models rather than traditional IT systems. For manufacturers, the consequences of compromised AI often surface as production disruptions, quality defects, or planning errors rather than obvious security incidents.
Compliance Gaps and the Supplier AI Blind Spot
The report also highlights growing exposure on the regulatory front. Only 15% of manufacturers conduct formal privacy impact assessments for AI, and just 19% maintain audit trails robust enough to demonstrate compliance. As AI oversight frameworks expand globally, including sector-specific expectations for transparency and accountability, these documentation gaps could become a material liability.
Compounding the issue is the accelerating convergence of operational technology and AI, which is moving faster than many IT-centric governance models can adapt. AI-driven decisions are increasingly embedded in shop-floor systems, logistics optimization tools, and supplier-facing platforms, often outside the direct control of central compliance teams.
Supply chain AI risk remains particularly under-governed. Many manufacturers rely on AI-enabled tools from suppliers and logistics partners without clear visibility into how those systems are trained, secured, or tested. When a supplier’s AI fails or is compromised, the impact is felt immediately on the production line or in delivery performance, not in policy reviews.
“Manufacturers have world-class supply chain discipline, but AI has entered the ecosystem faster than governance,” said Patrick Spencer, senior vice president of Americas marketing and industry research at Kiteworks, in an official statement. “When supplier AI systems fail, the impact shows up on the production line, not in a policy document.”
When Resilience Assumptions Quietly Break
One underappreciated risk is that many manufacturers still treat AI as a component to be hardened, rather than as a decision layer that reshapes how failure propagates. In traditional operations, redundancy absorbs shocks locally. With AI-driven planning, quality, and supplier coordination, errors or manipulation can replicate across sites and partners before they are detected. Recent industry research shows that adversarial testing and evidence trails are not merely security exercises but mechanisms for slowing that propagation. As AI systems increasingly arbitrate trade-offs between cost, speed, and risk, governance gaps stop being abstract weaknesses and start functioning as amplifiers of disruption. The operational discipline that manufacturers apply to physical processes may need to extend, with equal rigor, to how AI decisions are stress-tested, documented, and bounded, before those decisions harden into invisible dependencies.
