Proof over Promise Insights on Real World AI Adoption from 2025 MINDS Organizations 2026

Beyond data quality and technical limits, many MINDS organizations identify trust, reliability, accuracy, human oversight and compliance as core challenges. Sustainable AI adoption requires an ecosystem of well-designed principles, practices and controls – collectively referred to as “responsible AI” – to effectively govern the technology for desirable outcomes.5 As AI transforms more business processes, organizations must operationalize responsible AI at scale to ensure trust, resilience and meaningful human judgment where it matters most. Technical controls are being embedded into AI systems to cultivate trust and enable scalable governance Several MINDS organizations are shifting from policy- heavy oversight to technology-enabled governance, embedding responsible AI principles directly into the systems and operational workflows themselves. This approach to adaptive governance moves beyond static guidelines towards dynamic real-time enforcement of trust mechanisms (such as traceability, explainability and fairness) directly within the AI life cycle. By integrating controls like model monitoring, bias detection and secure data pipelines into composable AI platforms and agentic AI systems, organizations are preparing resilient infrastructures capable of scaling with responsibility in mind. For example, organizations like CATL and Deep Principle have implemented multi-tiered security systems and automated compliance checks that align with global AI governance frameworks and regulatory standards while maintaining human oversight where appropriate. These safeguards cut risk and speed deployment by embedding governance from the outset. This “trust-by-design” paradigm is fast becoming a foundation for enterprise-scale AI transformation. Human oversight is being right-sized for varying degrees of automated decision-making A more nuanced model of human oversight is emerging. Rather than inserting humans into every decision loop, organizations are calibrating oversight to the level of autonomy, risk and decision complexity, signalling a mature, context-aware approach to responsible AI. Across the MINDS cohorts, three governance archetypes are emerging: –Full autonomy with human override capabilities: In low-risk, well-bounded environments, AI systems are granted full autonomy with the option for human override. Siemens and Schneider Electric exemplify this model, using AI to autonomously optimize building temperatures. These systems act directly on the physical world, but the consequences of error are minimal and reversible, making light-touch oversight sufficient. –Bounded autonomy in structured contexts: In moderately complex scenarios, AI operates within predefined action spaces and structured environments. Lenovo and Fujitsu’s supply chain orchestration systems, and EXL’s coding assistants, fall into this category. Here, AI agents make decisions independently but within tightly scoped parameters, ensuring that governance is embedded through design constraints rather than constant human supervision. –Human-governed autonomy for high-stakes decisions: In high-risk or sensitive domains, human oversight remains essential. Whether it’s Ant Group’s diagnostic AI or State Grid Corporation of China’s grid management systems, these applications require human validation before AI outputs are acted upon. This tiered approach shows that the level of human involvement should be proportional to the potential impact of its decisions and the context within which they are made. Rather than defaulting to binary governance models, organizations appear to be experimenting with risk-calibrated approaches in which human roles are strategically designed to complement AI capabilities. 2.5 Insight 5: Scaling AI with confidence through responsible AI practices The ‘trust-by- design’ paradigm is fast becoming a foundation for enterprise-scale AI transformation. Proof over Promise: Insights on Real-World AI Adoption from 2025 MINDS Organizations 23