Adaptive AI for Detecting Modern DGA Attacks
Akamai Blog, Tuesday, July 7th, 2026
Akamai's hybrid deep learning model detects modern language-mimicking DGA malware domains with low false positives.
Akamai describes an adaptive AI approach to detecting modern domain generation algorithm (DGA) attacks, where malware produces domains mimicking natural language to evade blocklists and signature defenses.
It developed a hybrid CNN-BiLSTM-Attention deep learning framework combining local pattern extraction, sequential context modeling, and attention-based prioritization to catch both random and dictionary-based DGA domains. The model trained on roughly 20 million labeled domains (10 million legitimate, 10 million DGA) from threat intelligence feeds and internal DNS telemetry.
It outperformed baselines with notably low false-positive rates, reducing alert fatigue, and includes continuous learning that monitors data drift to trigger fine-tuning or full retraining as techniques evolve.