Advancing Responsible AI Innovation A Playbook 2025

Play 8 Scale responsible AI with technology enablement As AI applications multiply at pace and the risk landscape grows more complex, responsible AI technologies become indispensable – from operationalized platforms to systemic enablement and continuous oversight. Organization leaders Key roadblocks that arise within the organization Limited visibility into enterprise-wide AI usage and risks, impacts maintaining a systematic and comprehensive inventory capturing all assets in use Contending with technical debt from legacy technologies, exposing organizations to heightened AI risks and security vulnerabilities (which also hinder the systematic implementation of technologies designed to integrate trust and regulatory compliance into AI systems) Human review bottlenecks, preventing automation of risk assessments for AI use cases and resulting in less responsive processes Actions for organization leaders –Systematize responsible AI: Identify and use dedicated technology solutions that support the operationalization and scaling of responsible AI tasks, including for and with agentic AI systems (see Case study 9). Examples include: –Real-time monitoring: Multiple technologies can support continuous AI oversight. A control plane offers centralized governance across distributed systems, while monitoring tools, sensors and agents enable real-time tracking of system performance, security events and adherence to responsible AI and compliance metrics. –AI agents: These can support in analysing vast threat intelligence and delivering real-time assessments.87 They may also enhance risk management by scanning and evaluating AI outputs against responsible AI metrics and stress-testing models for alignment. –Red teaming: Efforts to proactively identify AI system vulnerabilities and ensure resiliency benefit from augmentation with embedded technology solutions to ensure evergreen testing against evolving risks. –Hardwire responsible AI controls into enterprise AI infrastructure and solutions: This incentivizes fluid adoption, accountability and decreases the likelihood of risks being overlooked. Employee upskilling initiatives to make use of responsible AI technologies within workflows may be needed (see Play 9) alongside upgrading legacy systems to a modern digital core. This includes integrating advanced data and AI management tools that support seamless and secure data and AI connectivity across the enterprise.88 –Maintain sufficient human oversight: To ensure accountability and offset limitations with AI, e.g. hallucinations and reasoning gaps or overreliance on AI outputs. The mandates and cadence of human oversight must adapt to increasingly autonomous and complex agentic AI systems and their potential for unintended consequences. There is an emerging market of platforms that help automate key steps, including AI system registration, risk assessment, requirements assignment and compliance sign- off, while supporting human oversight. Advancing Responsible AI Innovation: A Playbook 31