sponge-attack

Live1

1. live
   Chatter Attack Resource Consumption Technique for Large Language Models

   Research · since 10 Nov 2026

   Researchers demonstrated Chatter Attack, an attack technique that induces large language models into generating excessively long outputs during inference. The approach significantly drives up computational latency and operational hosting costs.

   Do:Implement strict inference timeout thresholds and maximum output token limits to limit exposure to resource consumption attacks.

   AI

Earlier0

Nothing closed yet.