The data foundation

Many of the most successful AI deployments in telecom share a common characteristic: trusted data foundations. When organizations discuss AI performance, attention often focuses on model selection, agent frameworks, and emerging technologies. Yet the quality of outcomes frequently depends on something less glamorous but far more important: the underlying data. Reliable data creates predictable outcomes. Structured data creates repeatable outcomes. Governed data creates auditable outcomes. In practice, organizations that have successfully introduced greater autonomy into fault management and assurance consistently find trusted data to be the primary enabler of success.

Conversely, poorly governed data introduces uncertainty. AI systems become harder to explain, harder to validate, and harder to trust. As organizations move toward autonomous operations, data governance is therefore no longer an administrative exercise. It becomes a strategic capability. The path to trusted AI begins long before model deployment. It begins with confidence in the information that powers decision-making.

The security imperative

Trust cannot be discussed without addressing security. Threat actors are already benefiting from AI. Cybercriminals are using automation and AI to accelerate reconnaissance, identify vulnerabilities, and increase the speed of attacks. The industry has long argued for a shift from human-speed to machine-speed security operations, and AI is increasingly making that possible.

But operationalizing AI-driven security in telecoms is not straightforward. Operators manage some of the world's most critical digital infrastructure, where changes are carefully controlled, and security architectures carry significant governance requirements. The challenge is not whether AI can improve security outcomes. It is how to integrate it safely without creating new vulnerabilities or undermining existing protections.

Technology may be advancing rapidly. Operational change rarely moves at the same pace.

The governance challenge of agentic AI

The emergence of agentic AI introduces another layer of complexity. Unlike traditional software applications, agents may make decisions, interact with systems, and collaborate with other agents to achieve objectives. As organizations deploy increasing numbers of these autonomous capabilities, governance becomes significantly more difficult. This is the same Pokémon challenge at enterprise scale: before organizations can govern agentic AI, they must first find, identify, and catalog every moving part.

The questions are fundamental: how many agents exist across the organization, what models are they using, what data can they access, which decisions are they authorized to make, and how are they being monitored? These are governance questions, but they are also business questions. The answers influence security, compliance, operational resilience, and customer trust. They also influence costs. As AI usage grows, organizations need greater transparency into how models are consumed, where workloads are processed, and how resources are allocated.

The industry therefore faces a new requirement: managing AI ecosystems rather than individual AI solutions. Success will depend less on deploying the latest model and more on establishing effective mechanisms for governance, orchestration and oversight.

Closing the gap

The scale of the challenge is measurable. Published industry research surveying 130 operators worldwide found that 72 percent of communication service providers believe their AI is trustworthy, but only 14 percent can produce evidence to prove it. Governance frameworks exist on paper. Operational proof does not. Closing that gap requires progress on three fronts. 

Organizations need a standardized approach to sourcing, operating, and governing AI models. Whether models come from hyperscalers, vendors, or in-house teams, a common control layer is what makes model behavior consistent, auditable, and lifecycle-managed.