Challenges and mitigations of GenAI for networking

Written by Alejandro Medina, CTO
Published by The Fast Mode on 6 March 2025

Generative AI (GenAI) is transforming networking by automating operations, enhancing security and optimising performance. It enables networks to become self-sufficient, reducing manual intervention while improving efficiency.

A key innovation in this field is intent-based networking (IBN), where AI translates business objectives into automated network policies. This ensures networks operate as intended while adapting to changing conditions. GenAI enhances intent operations by predicting and resolving issues proactively, improving reliability and performance.

Challenges

Despite its advantages, GenAI faces challenges, particularly biases and inaccuracies in AI-driven networking models. AI learns from historical data, and if that data is flawed or unrepresentative, the model may inherit and amplify biases.

For example, an AI-powered intrusion detection system trained only on specific attack patterns may fail to detect new or regional threats. Similarly, biased traffic management algorithms could cause uneven bandwidth allocation, reducing performance for certain users.

These issues arise due to poor quality data, lack of validation and insufficient diversity in training sets. If unaddressed, they can undermine trust in AI-driven networking.

Mitigations

Data quality enhancements

Reliable AI models require high-quality, diverse, and up-to-date data to reduce bias and improve accuracy.

• Diverse data sources – Training AI on datasets from various network environments prevents bias. For instance, an AI model trained only on corporate networks may struggle with cloud-based infrastructures
• Removing outdated and faulty data – AI models must continuously update their knowledge. An outdated model relying on old cybersecurity data may fail to detect new threats
• Holistic data integration – Combining real-time network logs, historical data and external threat intelligence provides a comprehensive view of network conditions, improving decision-making

Adaptive learning

AI models should continuously refine themselves to stay relevant. Adaptive learning ensures AI remains accurate as network conditions evolve.

• Real-time feedback loops – AI systems should adjust based on past performance. If a traffic management AI causes congestion, it should learn and optimise its routing
• Self-learning threat detection – AI security systems should detect new cyber threats dynamically rather than relying solely on predefined attack signatures
• Human-in-the-loop AI – While AI automates tasks, human oversight ensures models remain fair and effective, balancing automation with expertise

Conclusion

GenAI is revolutionising networking, but bias and inaccuracy can limit its effectiveness. By improving data quality and adaptive learning, organisations can maximise AI’s benefits while ensuring fairness and reliability. The future of AI-driven networking depends on responsible development and continuous refinement.