Distributed Denial of Service attacks have evolved into persistent, large-scale threats to Internet availability, with amplification-based techniques enabling adversaries to generate terabit-per-second traffic volumes by exploiting protocol asymmetries. Traditional mitigation strategies, often relying on static filtering rules or control-plane-based inspection, struggle to cope with increasingly dynamic and adaptive attack patterns. As a result, recent research has progressively shifted toward In-Network Machine Learning, leveraging programmable data planes enabled by the P4 language to detect and mitigate attacks directly within network switches at line rate. Programmable switches allow customized packet parsing, telemetry extraction, and stateful processing inside the forwarding pipeline, making it possible to embed security mechanisms directly into network hardware.