LLMs exploited to boost real-time ASR adversarial attacks

Researchers have developed a new method called Semantic Gambit that uses large language models to enhance acoustic adversarial attacks against automatic speech recognition (ASR) systems. By leveraging the predictive capabilities of LLMs, this attack overcomes the temporal constraints of real-time ASR, leading to a significant increase in word error rates. The findings highlight a potential vulnerability in common, low-latency LLM tools that could be exploited to compromise real-time ASR pipelines. AI

IMPACT Demonstrates a new vulnerability in real-time ASR systems, potentially impacting the security of voice-enabled applications.

RANK_REASON The cluster contains a research paper detailing a novel attack method. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Large Language Model
• Semantic Gambit

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →