AI governance for the agentic AI era
Artificial intelligence has entered a bold new era characterized by autonomous systems, known as agentic AI, that can perceive, reason, plan, and act with minimal human involvement. These intelligent agents represent a significant breakthrough in AI-driven decision-making, allowing enterprises to automate complex workflows and adapt in real time. However, as agentic AI systems take on complex, high-value roles, they introduce a spectrum of risks due to their independent learning, reasoning, and action capabilities. Trust becomes central in harnessing the full potential of agentic AI by implementing systems that are intelligent, accountable, transparent, and aligned with human values.
Understanding agentic AI through the KPMG TACO FrameworkTM
Agentic AI represents a significant evolution beyond traditional AI. While conventional AI excels at classification, prediction, and pattern recognition, it remains static, requiring periodic retraining and human oversight.
As AI agentic systems proliferate and scale, a structured framework is essential to understand and categorize them based on their capabilities.
To make sense of these variations, we created the KPMG TACO Framework – which classifies agents into four key types: Taskers, Automators, Collaborators, and Orchestrators. Each of the types leverages the same foundational tools and capabilities – goal interpretation, reasoning engines (using advanced models including LLMs), memory, tools, and orchestration – but differ in goal planning, execution and complexity.
The four main types of AI agents
1
2
3
4
Mitigating agentic AI risks with the KPMG Trusted AI Framework™
As we push the boundaries of what AI can do, we must confront not only technical challenges, but also deep ethical and societal questions about control, accountability, and trust.
The KPMG Trusted AI framework offers an actionable approach to managing these risks in a responsible and ethical manner. It equips organizations with the tools to embed ethical principles and governance into every stage of the AI lifecycle—from design and deployment to monitoring and evolution. Below are key highlights of how this framework can be applied to help ensure trusted agentic AI deployments.
- Reliability: Refers to the extent to which AI systems perform consistently with their intended purpose, scope, and required level of precision. Reliability requires not only robust design and testing but also continuous monitoring to align outcomes with human expectations and values.
- Accountability: Helps ensure that organizations clearly define responsibility for AI-driven decisions and that there is an audit trail of agent activity.
- Transparency & Explainability: Transparency refers to the ability to understand how and why an AI system functions, while explainability focuses on making its decisions interpretable to humans.
- Security & Safety: Security involves implementing robust and resilient practices to protect AI systems from unauthorized access, manipulation, or disruption, while safety focuses on preventing emotional and/or physical harm to people, businesses, and property.
- Data Privacy: By promoting strong data governance practices, including the use of anonymized and ethically sourced data, strict access controls, and compliance with data protection regulations. These safeguards help ensure that even as data volumes grow, individual privacy remains protected.
- Fairness: Stresses the importance of limiting bias against individuals, communities or groups. To address these risks, the framework encourages organizations to embed fairness metrics and thresholds into agent design, data sources and governance, supported by continuous evaluation and feedback mechanisms.
Dive into our thinking:
AI governance for the agentic AI era
Download nowThe first 10: Building a trusted foundation for agentic AI
Aligned with the KPMG Trusted AI framework, these top 10 control considerations serve as a powerful foundation for deploying agentic AI responsibly and effectively. This is not a one-time checklist—it’s a continuous, evolving journey.Each step forward strengthens trust, transparency, and resilience.
- Assess agent risk
- Determine human oversight requirements
- Establish default scope boundaries
- Reveal agent’s chain-of-thought thinking
- Assign unique identifiers for attributability
- Design immutable logging and monitoring
- Design multi-agent systems to help prevent cascading failures
- Build fail-safe and fallback protocols
- Deploy AI red-teaming
- Ongoing evaluations & feedback
Conclusion
The agentic wave marks a pivotal moment in business transformation, creating a new playing field where autonomous systems drive unprecedented value and competitive advantage. Success demands more than technological prowess; it requires a deliberate approach that places trust at the core of AI deployment. Organizations must build their agentic AI foundations on robust governance frameworks for innovation and responsible implementation.
The KPMG Trusted AI framework provides the essential foundation, enabling businesses to accelerate transformation while maintaining control and building stakeholder trust. Are you ready to lead with agentic AI? The future is here—dynamic, powerful, and full of promise. Seize it.
Explore more
Insight
The agentic AI advantage: Unlocking the next level of AI value
Learn how agentic AI can unlock enterprise value, supercharge adoption, boost workforce productivity, and revolutionize process engineering.
KPMG Workbench
Help supercharge your business with this multi-agent AI platform—combining advanced, trusted AI agents with the insight and deep industry expertise of KPMG professionals
Meet our team
