The telecom industry is racing to build trust into artificial intelligence systems that will soon make autonomous decisions in critical networks, driven by new regulations and the shift from AI as a helpful tool to AI as a decision-maker. As networks become more autonomous, the stakes of AI failures grow exponentially, forcing operators and equipment makers to rethink how they design, test, and deploy AI in live systems.

Why Does Trust Matter More Now Than Ever in Telecom AI?

For decades, AI in telecom networks played a supporting role, helping engineers optimize traffic or predict equipment failures. But as 5G and 6G networks evolve, AI is shifting from an assistive tool to a decisive system that makes real-time decisions affecting millions of users. This transition introduces a fundamental paradox: while AI is essential for managing the scale and complexity of modern networks, its opaque decision-making and vulnerability to attacks pose serious risks to safety, security, and transparency.

The regulatory landscape is tightening around this shift. The European Union's AI Act, which came into force in 2024, applies a risk-based approach with stricter obligations for higher-risk systems. For telecom operators deploying AI in live networks, compliance is no longer optional, it is a market requirement.

Ericsson, one of the world's largest telecom equipment makers, has adopted the EU's AI ethics guidelines since 2019 and now emphasizes that trustworthiness must be embedded by design in telecom systems. The company stated that for AI to be integrated into the telecom domain, including 5G and 6G networks, it must move beyond mere performance metrics to a holistic framework of trustworthiness.

What Does "Trustworthy AI" Actually Mean in Practice?

Trustworthiness in telecom AI encompasses five interconnected dimensions: safety, security, transparency, reliability, and ethics. Each dimension addresses a specific vulnerability in autonomous systems. Safety ensures the system works as intended and does no harm. Security protects against attacks that could compromise network integrity. Transparency allows engineers and customers to understand why AI made a particular decision. Reliability ensures consistent performance under real-world conditions. Ethics ensures the system respects human values and avoids discriminatory outcomes.

One of the most critical techniques for building transparency is explainable AI, or XAI. Rather than treating AI models as black boxes, XAI provides interpretable reasoning for why a decision was made. At the model level, XAI helps developers understand internal logic and validate correctness. At the system level, it traces the entire flow from inputs through the model to final outcomes, enabling root-cause analysis and proactive issue detection before problems appear in live networks.

How to Strengthen AI Governance in Telecom Networks

• Data Quality and Provenance: ML models derive their behavior from training data, so corrupted or biased data directly affects model outputs. Organizations must analyze data for correctness, bias, and representation of the entire input domain, track data provenance to prevent tampering, and retain training data for forensic analysis in case of incidents.
• Privacy by Design: Sensitive data such as personally identifiable information (PII) requires de-identification, but even de-identified data can be vulnerable to extraction attacks if models are queried with external data sources. Privacy principles must be embedded throughout the model lifecycle, not added afterward.
• Defense Against Adversarial Attacks: Attackers can use data poisoning to introduce bias into models, adversarial examples to change outputs with undetectable changes to input data, or extraction attacks to steal model weights. Layered defenses and continuous monitoring are essential to detect and mitigate these threats.
• Explainability at Multiple Levels: Both external users such as customers and internal users such as developers and testers benefit from explainability insights. System-level explainability helps automated AI lifecycle management through compositional, generalized explanations that can be understood by non-experts.

Beyond individual organizations, industry coalitions are emerging to set standards and share best practices. The Cloud Security Alliance (CSA) launched its AI Safety Initiative to develop vendor-neutral guidance and tools that empower organizations of all sizes to deploy AI solutions that are safe, responsible, and compliant. The initiative includes an updated AI Controls Matrix with 247 control objectives across 18 security domains, along with implementation guidance, auditing resources, regulatory mappings, and assessment tools.

The CSA's approach reflects a broader recognition that AI governance cannot be left to individual companies. The organization convenes an Executive Leadership Council composed of government agencies, industry titans, and innovators to establish best practices and a "north star" for responsible AI deployment.

What Happens When AI Systems Fail in Critical Infrastructure?

The stakes of inadequate AI governance in telecom are extraordinarily high. A single misconfigured AI model in a 5G network could disrupt service for millions of users, compromise network security, or enable attackers to manipulate traffic routing. Unlike traditional software bugs that affect a limited set of users, AI failures can cascade unpredictably across entire network architectures because AI systems interact with and influence one another in ways that are difficult to predict.

This risk is why Ericsson and other operators emphasize that trustworthiness is now a must for compliance and market confidence. Operators and subscribers expect it, and competitors will offer it. Organizations that fail to embed trustworthiness by design risk regulatory penalties, customer loss, and reputational damage.

The convergence of autonomous networks, generative AI, and agentic systems (AI systems that can act independently to achieve goals) introduces new challenges that traditional AI governance frameworks were not designed to address. Large language models (LLMs) can interpret and generate natural language, reason about complex scenarios, and operate autonomously in the real world. This capability makes trustworthiness a moving target, requiring continuous evolution of techniques and oversight mechanisms.

As networks progress toward higher levels of autonomy, AI and machine learning functions are being integrated across mobile network architecture, from lower-level layers to high-level management. Ensuring these AI entities can be trusted is no longer a nice-to-have feature, it is a foundational requirement for network reliability and regulatory compliance.