Operation visibility and understanding: The shared foundation of AI trust and performance

In case anyone needed reminding, the summer’s Hugging Face “security incident” illustrated the vital importance of AI governance in the agentic era. As AI development moves rapidly into ever more diverse use cases, regulators have struggled to keep up with how and where legislative oversight should be imposed.

But European authorities have stepped up. With the EU AI Act’s general application date arriving in August 2026, companies must now make AI use more transparent, prevent prohibited uses, improve workforce competence, and prepare evidence of responsible governance, with high-risk AI system obligations to follow over 2027-28.

Europe is not regulating AI because it wants to slow innovation; it is trying to make AI trustworthy at scale. But trust requires evidence, and evidence requires visibility. The EU AI Act marks an important shift from AI principles to enforceable accountability. For organizations, compliance will depend not only on documenting which models they use, but on being able to trace AI decisions in practice.

What’s more, this operational visibility will be as critical to extracting maximum value from AI systems as it is for regulatory compliance, particularly when AI agents are used.

The trust gap

As AI agents shoulder more operational responsibility, the need for traceability and transparency regarding the drivers of agentic decision-making is becoming increasingly urgent.

While many companies use sophisticated technology to monitor people, applications, and even infrastructure, there’s an AI agent-shaped hole in their oversight. Many businesses have little or no capability to monitor agents. This means they have limited understanding of how agents interact with business processes, where they deviate from intended behavior, and whether their actions align with organizational policies.

The unintended hack of Hugging Face by agents that exploited a previously unknown vulnerability in a testing sandbox environment is now the defining example of this agentic gap. While looking for solutions to a cybersecurity benchmarking test, the rogue AI agents reportedly collaborated for weeks to escape containment. They even ‘chatted’ about their plans on an improvised message board, unnoticed by the very organisation that deployed them.

The trust and traceability issue is putting the brakes on successful enterprise AI rollout. According to PwC’s 2025 Responsible AI survey, only half of respondents reported being very effective at tracking and inventorying their AI use cases. And in the closely regulated financial services sector, managing AI risks (29%) and meeting regulatory obligations (28%) are among the leading barriers preventing organisations from scaling AI successfully, according to Deloitte.

AI systems and agents are only as valuable as they are auditable. Without tight governance over their reasoning and actions, transformative power becomes a high-risk liability.

The key to AI compliance and performance

The ability to trace AI decisions is far more than a compliance concern, it is also the root of Enterprise AI success. To keep AI systems optimized and aligned to evolving needs, it’s essential to understand and update the data foundation and business rules that enable the platforms.

Increasingly this makes AI governance and performance a function of its ability to see and understand an organization’s operational reality. After all, you can neither fix nor govern what you can’t see and do not understand. When an AI agent operates across fragmented enterprise systems, a flawed credit decision, a supply chain disruption, or a procurement error cannot be properly understood. Not without visibility into and understanding of the full chain of execution that led to that outcome. The question regulators and businesses will increasingly need to answer is not just ‘what model was used?’ but ‘why did this decision happen, and can we demonstrate appropriate oversight?’.

However, operational visibility and understanding are often obscured by fragmented legacy tech stacks, disconnected data platforms, low data quality, and siloed functional processes. For example, 87% of respondents to PwC’s 2026 Digital Trends in Operations Survey say poor data quality has impacted their organization’s ability to achieve value for digital initiatives. Similarly, 42% cite siloed organizational structures and processes as among the biggest barriers to achieving the horizontal, networked operating model needed for cross-functional insight and end-to-end visibility.

Addressing these issues starts with generating an accurate, real-time picture of the enterprise as one connected, dynamic system rather than patchwork of disconnected parts.

Enabling operational visibility and understanding, unlocking AI trust

At Celonis, our approach centers on turning AI’s operational blind spots into operational clarity.

The Celonis Platform brings together process data, business knowledge, and intelligence from across an organization's systems, applications, and interactions to show how the organization operates. It creates a living digital twin of business operations that AI can understand, so that agentic solutions know what’s happened in the past, what’s happening now, and what should happen in the future. The platform helps organizations deploy AI where it can have the greatest impact. And, it enables AI solutions to be governed, monitored, optimized and trusted with business-critical operations.

Read more: Celonis Named a Leader in 2026 Gartner® Magic Quadrant™ for Digital Twin of an Organization Platforms

AI understanding is currency

The more sophisticated AI becomes, and the more autonomy it's given, the more critical it is to understand how it makes decisions. That's not only a compliance question, although the EU AI Act's fines (up to €35 million or 7% of global turnover) are motivating enough on their own.

It's also a question of Enterprise AI success and business continuity. Without operational visibility, businesses cannot steer AI performance toward their strategic goals. Worse, they're ceding critical decisions to a system they can neither fully explain nor hold accountable.

That makes operational visibility a strategic priority in its own right—the catalyst for AI that is optimized, auditable, and genuinely trustworthy.