Scholarlink

<p>The Spectre attack exploits speculative execution in modern processors, allowing attackers to leak sensitive information by manipulating branch prediction and executing instructions along speculative paths. The spectre attacks utilize cache timing analysis as a side channel to infer protected memory contents, bypassing traditional security boundaries. To address the challenge, the paper proposes a novel framework that integrates Conditional Generative Adversarial Networks (CGANs) for synthetic data augmentation and Long Short-Term Memory (LSTM) networks for real-time anomaly detection. CGANs generate diverse speculative execution scenarios, enhancing the dataset and improving model robustness against new Spectre variants. LSTM models analyze temporal patterns in Hardware Performance Counters (HPCs) to detect subtle deviations in CPU behavior indicative of Spectre exploitation. The proposed CGAN-LSTM framework dynamically adapts to evolving attack patterns, achieving a 31% improvement in detection accuracy over traditional methods while maintaining a low false-positive rate. Experimental results demonstrate the model’s effectiveness in detecting both known and previously unseen Spectre variants in real-time. The proposed work highlights the potential of combining machine learning with advanced speculative execution analysis to mitigate sophisticated side-channel vulnerabilities in modern computing systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing spectre attack detection with conditional GAN data augmentation and LSTM classification

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Bhukya Krishna Priya,
  • Jobin Jose,
  • Payarda Santosh Babu,
  • Ruhul Amin

摘要

The Spectre attack exploits speculative execution in modern processors, allowing attackers to leak sensitive information by manipulating branch prediction and executing instructions along speculative paths. The spectre attacks utilize cache timing analysis as a side channel to infer protected memory contents, bypassing traditional security boundaries. To address the challenge, the paper proposes a novel framework that integrates Conditional Generative Adversarial Networks (CGANs) for synthetic data augmentation and Long Short-Term Memory (LSTM) networks for real-time anomaly detection. CGANs generate diverse speculative execution scenarios, enhancing the dataset and improving model robustness against new Spectre variants. LSTM models analyze temporal patterns in Hardware Performance Counters (HPCs) to detect subtle deviations in CPU behavior indicative of Spectre exploitation. The proposed CGAN-LSTM framework dynamically adapts to evolving attack patterns, achieving a 31% improvement in detection accuracy over traditional methods while maintaining a low false-positive rate. Experimental results demonstrate the model’s effectiveness in detecting both known and previously unseen Spectre variants in real-time. The proposed work highlights the potential of combining machine learning with advanced speculative execution analysis to mitigate sophisticated side-channel vulnerabilities in modern computing systems.