Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Learn about a novel cybersecurity attack methodology that exploits memory access patterns in confidential computing environments through generative language modeling techniques. Discover how researchers from Yale University developed FIT (Found in Translation), an innovative attack approach that leverages correlations in page access patterns to reconstruct sensitive application data with unprecedented accuracy. Explore the parallels between application page access patterns and natural language grammatical structures, and understand how this insight enables the use of recurrent encoder-decoder architectures to predict application-level object accesses from page-level access sequences. Examine the vulnerabilities in cloud provider-controlled Operating System stacks that manage memory services like paging, and see how these can be exploited even in hardware-protected confidential computing environments. Analyze evaluation results demonstrating FIT's effectiveness against popular AI/ML model inference services and semantic search applications, achieving prediction accuracies ranging from 71.7% to 99.9% while significantly outperforming existing state-of-the-art memory access pattern attacks.
