Pramod Bhatotia

Dimitrios Stavrakakis, Masanori Misono, Julian Pritzi et al.

Privacy regulations such as the General Data Protection Regulation (GDPR) impose strict requirements on how personal data is stored, processed, and audited. While key-value stores (KVS) are widely used in latency-sensitive applications, their simple data model and untrusted cloud deployment environments make GDPR compliance particularly challenging. Existing approaches require invasive code modifications, impose high performance overheads, or overlook the integrity of compliance mechanisms themselves. This paper presents GDPRuler, a trusted middleware system that enables verifiable GDPR compliance for KVS on untrusted clouds without modifying their codebase. GDPRuler deploys a trusted GDPR monitor inside a Confidential Virtual Machine (CVM), which enforces GDPR policies, manages compliance metadata, and maintains tamper-evident audit logs. A declarative policy language translates core GDPR obligations into enforceable runtime rules. To ensure efficiency, GDPRuler encodes metadata compactly within KV records, builds dedicated metadata indexes for GDPR-specific queries, and logs only compliance-relevant events in a space-efficient format. We implement GDPRuler as a transparent proxy for unmodified Redis and RocksDB deployments. Evaluation with YCSB and GDPR-inspired workloads shows that GDPRuler enforces core compliance guarantees with low overheads: GDPRuler achieves ~61% of native KVS throughput with the CVM environment contributing 28%-32% of it, metadata storage overhead remains below 20%, and GDPR queries benefit from 13-182x speedup through metadata indexing. By embedding verifiable policy enforcement into a trusted middleware layer, GDPRuler offers a practical path toward GDPR-compliant KVS on untrusted cloud infrastructures.

Teofil Bodea, Masanori Misono, Julian Pritzi et al.

AI agents powered by large language models are increasingly deployed as cloud services that autonomously access sensitive data, invoke external tools, and interact with other agents. However, these agents run within a complex multi-party ecosystem, where untrusted components can lead to data leakage, tampering, or unintended behavior. Existing Confidential Virtual Machines (CVMs) provide only per binary protection and offer no guarantees for cross-principal trust, accelerator-level isolation, or supervised agent behavior. We present Omega, a system that enables trusted AI agents by enforcing end-to-end isolation, establishing verifiable trust across all contributing principals, and supervising every external interaction with accountable provenance. Omega builds on Confidential VMs and Confidential GPUs to create a Trusted Agent Platform that hosts many agents within a single CVM using nested isolation. It also provides efficient multi-agent orchestration with cross-principal trust establishment via differential attestation, and a policy specification and enforcement framework that governs data access, tool usage, and inter-agent communication for data protection and regulatory compliance. Implemented on AMD SEV-SNP and NVIDIA H100, Omega fully secures agent state across CVM-GPU, and achieves high performance while enabling high-density, policy-compliant multi-agent deployments at cloud scale.

Do Le Quoc, Franz Gregor, Sergei Arnautov et al.

Data-driven intelligent applications in modern online services have become ubiquitous. These applications are usually hosted in the untrusted cloud computing infrastructure. This poses significant security risks since these applications rely on applying machine learning algorithms on large datasets which may contain private and sensitive information. To tackle this challenge, we designed secureTF, a distributed secure machine learning framework based on Tensorflow for the untrusted cloud infrastructure. secureTF is a generic platform to support unmodified TensorFlow applications, while providing end-to-end security for the input data, ML model, and application code. secureTF is built from ground-up based on the security properties provided by Trusted Execution Environments (TEEs). However, it extends the trust of a volatile memory region (or secure enclave) provided by the single node TEE to secure a distributed infrastructure required for supporting unmodified stateful machine learning applications running in the cloud. The paper reports on our experiences about the system design choices and the system deployment in production use-cases. We conclude with the lessons learned based on the limitations of our commercially available platform, and discuss open research problems for the future work.

Bohdan Trach, Rasha Faqeh, Oleksii Oleksenko et al.

A lease is an important primitive for building distributed protocols, and it is ubiquitously employed in distributed systems. However, the scope of the classic lease abstraction is restricted to the trusted computing infrastructure. Unfortunately, this important primitive cannot be employed in the untrusted computing infrastructure because the trusted execution environments (TEEs) do not provide a trusted time source. In the untrusted environment, an adversary can easily manipulate the system clock to violate the correctness properties of lease-based systems. We tackle this problem by introducing trusted lease -- a lease that maintains its correctness properties even in the presence of a clock-manipulating attacker. To achieve these properties, we follow a "trust but verify" approach for an untrusted timer, and transform it into a trusted timing primitive by leveraging two hardware-assisted ISA extensions (Intel TSX and SGX) available in commodity CPUs. We provide a design and implementation of trusted lease in a system called T-Lease -- the first trusted lease system that achieves high security, performance, and precision. For the application developers, T-Lease exposes an easy-to-use generic APIs that facilitate its usage to build a wide range of distributed protocols.

Sheung Chi Chan, James Cheney, Pramod Bhatotia et al.

System level provenance is of widespread interest for applications such as security enforcement and information protection. However, testing the correctness or completeness of provenance capture tools is challenging and currently done manually. In some cases there is not even a clear consensus about what behavior is correct. We present an automated tool, ProvMark, that uses an existing provenance system as a black box and reliably identifies the provenance graph structure recorded for a given activity, by a reduction to subgraph isomorphism problems handled by an external solver. ProvMark is a beginning step in the much needed area of testing and comparing the expressiveness of provenance systems. We demonstrate ProvMark's usefuless in comparing three capture systems with different architectures and distinct design philosophies.

Roland Kunkel, Do Le Quoc, Franz Gregor et al.

Machine learning has become a critical component of modern data-driven online services. Typically, the training phase of machine learning techniques requires to process large-scale datasets which may contain private and sensitive information of customers. This imposes significant security risks since modern online services rely on cloud computing to store and process the sensitive data. In the untrusted computing infrastructure, security is becoming a paramount concern since the customers need to trust the thirdparty cloud provider. Unfortunately, this trust has been violated multiple times in the past. To overcome the potential security risks in the cloud, we answer the following research question: how to enable secure executions of machine learning computations in the untrusted infrastructure? To achieve this goal, we propose a hardware-assisted approach based on Trusted Execution Environments (TEEs), specifically Intel SGX, to enable secure execution of the machine learning computations over the private and sensitive datasets. More specifically, we propose a generic and secure machine learning framework based on Tensorflow, which enables secure execution of existing applications on the commodity untrusted infrastructure. In particular, we have built our system called TensorSCONE from ground-up by integrating TensorFlow with SCONE, a shielded execution framework based on Intel SGX. The main challenge of this work is to overcome the architectural limitations of Intel SGX in the context of building a secure TensorFlow system. Our evaluation shows that we achieve reasonable performance overheads while providing strong security properties with low TCB.

Bohdan Trach, Alfred Krohmer, Sergei Arnautov et al.

Cloud computing offers the economies of scale for computational resources with the ease of management, elasticity, and fault tolerance. To take advantage of these benefits, many enterprises are contemplating to outsource the middlebox processing services in the cloud. However, middleboxes that process confidential and private data cannot be securely deployed in the untrusted environment of the cloud. To securely outsource middleboxes to the cloud, the state-of-the-art systems advocate network processing over the encrypted traffic. Unfortunately, these systems support only restrictive middlebox functionalities, and incur prohibitively high overheads due to the complex computations involved over the encrypted traffic. This motivated the design of Slick --- a secure middlebox framework for deploying high-performance Network Functions (NFs) on untrusted commodity servers. Slick exposes a generic interface based on Click to design and implement a wide-range of NFs using its out-of-the box elements and C++ extensions. Slick leverages SCONE (a shielded execution framework based on Intel SGX) and DPDK to securely process confidential data at line rate. More specifically, Slick provides hardware-assisted memory protection, and configuration and attestation service for seamless and verifiable deployment of middleboxes. We have also added several new features for commonly required functionalities: new specialized Click elements for secure packet processing, secure shared memory packet transfer for NFs chaining, secure state persistence, an efficient on-NIC timer for SGX enclaves, and memory safety against DPDK-specific Iago attacks. Furthermore, we have implemented several SGX-specific optimizations in Slick. Our evaluation shows that Slick achieves near-native throughput and latency.

Oleksii Oleksenko, Dmitrii Kuvaiskii, Pramod Bhatotia et al.

Memory-safety violations are a prevalent cause of both reliability and security vulnerabilities in systems software written in unsafe languages like C/C++. Unfortunately, all the existing software-based solutions to this problem exhibit high performance overheads preventing them from wide adoption in production runs. To address this issue, Intel recently released a new ISA extension - Memory Protection Extensions (Intel MPX), a hardware-assisted full-stack solution to protect against memory safety violations. In this work, we perform an exhaustive study of the Intel MPX architecture to understand its advantages and caveats. We base our study along three dimensions: (a) performance overheads, (b) security guarantees, and (c) usability issues. To put our results in perspective, we compare Intel MPX with three prominent software-based approaches: (1) trip-wire - AddressSanitizer, (2) object-based - SAFECode, and (3) pointer-based - SoftBound. Our main conclusion is that Intel MPX is a promising technique that is not yet practical for widespread adoption. Intel MPX's performance overheads are still high (roughly 50% on average), and the supporting infrastructure has bugs which may cause compilation or runtime errors. Moreover, we showcase the design limitations of Intel MPX: it cannot detect temporal errors, may have false positives and false negatives in multithreaded code, and its restrictions on memory layout require substantial code changes for some programs.

Do Le Quoc, Martin Beck, Pramod Bhatotia et al.

How to preserve users' privacy while supporting high-utility analytics for low-latency stream processing? To answer this question: we describe the design, implementation, and evaluation of PRIVAPPROX, a data analytics system for privacy-preserving stream processing. PRIVAPPROX provides three properties: (i) Privacy: zero-knowledge privacy guarantees for users, a privacy bound tighter than the state-of-the-art differential privacy; (ii) Utility: an interface for data analysts to systematically explore the trade-offs between the output accuracy (with error-estimation) and query execution budget; (iii) Latency: near real-time stream processing based on a scalable "synchronization-free" distributed architecture. The key idea behind our approach is to marry two existing techniques together: namely, sampling (used in the context of approximate computing) and randomized response (used in the context of privacy-preserving analytics). The resulting marriage is complementary - it achieves stronger privacy guarantees and also improves performance, a necessary ingredient for achieving low-latency stream analytics.