arXiv subjects

Prateek Saxena

Publications and source records attributed to Prateek Saxena.

At least 19 recordsLinked to original sources

arXiv2026

Mostly Automatic Translation of Language Interpreters from C to Safe Rust

Translating C programs to safe Rust is challenging owing to significant differences in typing constraints, ownership, and borrowing rules. Interpreter programs are particularly important targets for such translation, as they often handle untrusted inputs and suffer from memory-related vulnerabilities. We present Reboot, a mostly-automatic technique that translates real-world interpreter programs from C to safe Rust. Using Reboot, we have translated six interpreters ranging from 6k to 23k lines of C code to safe Rust, with each translation requiring only 1 to 11 brief user interventions. All translations pass 100% of the provided test suites, and achieve 62%--92% pass rates on separately created validation tests that were never exposed to the system. A security case study on mujs shows that memory vulnerabilities such as heap buffer overflows and use-after-free present in C are eliminated in the safe Rust translation. Two ideas underpin Reboot. First, feature reduction decomposes the translation by program features, creating a sequence of milestones where each is a complete, testable program; the translation starts from the simplest version and incrementally restores features, with each milestone validated before proceeding. Second, a multi-agent architecture orchestrates inherently unreliable coding agents through automated validation and feedback, keeping long-running translation workflows on track with minimal human involvement. An ablation study confirms that feature reduction improves translation correctness compared to using multi-agent translation alone, with 6%--20% improvements in pass rates on validation test suites.

cs.PL↗

arXiv2025

Epistemic Bias Injection: Manipulating LLM Opinion via Selective Context Retrieval

When answering user queries, LLMs often retrieve knowledge from external sources stored in retrieval-augmented generation (RAG) databases. These are often populated from unvetted sources, e.g. the open web, and can contain maliciously crafted data. This paper studies attacks that can manipulate the context retrieved by LLMs from such RAG databases. Prior work on such context manipulation primarily injects false or toxic content, which can often be detected by fact-checking or linguistic analysis. A more subtle threat, which we call epistemic bias injection (EBI), is where adversaries inject factually correct yet epistemically biased passages that systematically favor one side of an open-ended issue. Although linguistically coherent and truthful, such adversarial passages effectively crowd out alternative viewpoints during retrieval from the RAG and push LLM outputs towards an attack-desired stance. As a core contribution, we propose a novel characterization of the problem: We give a geometric metric that quantifies stance polarity and epistemic bias. This metric can be computed directly on embeddings of text passages. Leveraging it, we construct EBI attacks and develop a lightweight prototype defense called BiasDef for them. We evaluate them both on a comprehensive benchmark constructed from public question answering datasets. Our results show that: (1) the proposed attack induces significant stance polarity shifts, effectively evading existing retrieval-based sanitization defenses, and (2) BiasDef substantially reduces adversarial retrieval and epistemic bias in LLM's answers. Overall, this demonstrates the new threat as well as the ease of employing epistemic bias metrics for filtering in RAG-enabled LLMs.

cs.CR↗

arXiv2025

Model Inversion meets Cryptographic Fuzzy Extractors

Model inversion attacks pose an open challenge to privacy-sensitive applications that use machine learning (ML) models. For example, face authentication systems use modern ML models to compute embedding vectors from face images of the enrolled users and store them. If leaked, inversion attacks can accurately reconstruct user faces from the leaked vectors. A fuzzy extractor (FE) is a cryptographic primitive with properties that can help defend against model inversion, offering attack-agnostic security without requiring any re-training of the ML model it protects. To date, no systematic cryptanalysis of existing FE schemes that tolerate $\ell_2$ noise, as needed in modern ML-based face recognition systems, has been conducted. We perform the first in-depth security analysis of existing $\ell_2$-FE schemes showing that they offer weak security. We also show end-to-end inversion attacks that achieve high success rates in recovering original faces that are meant to be protected by FE schemes. We then offer a simple but new candidate scheme and prove its security formally. Our construction offers the first design point that offers practical runtime, stronger security, and usable accuracy for use in commodity ML-based face authentication.

cs.CR↗

arXiv2025

Adversarial Agent Collaboration for Correctness Improvements of C to Safe Rust Translation

Translating C to memory-safe languages, like Rust, prevents critical memory safety vulnerabilities that are prevalent in legacy C software. Even with recent LLM-based and tool-augmented translators, the resulting Rust code frequently diverges from the C source on inputs absent from the test suite used during translation; this correctness gap on unseen inputs remains a dominant obstacle to reliable, automatic C-to-Rust translation. In this work, we present ACToR (Adversarial C To Rust), a simple LLM-agent loop that closes this gap by adversarially searching for inputs on which the translation diverges from the C source, and using them to drive subsequent refinements. Inspired by GANs, ACToR pits a translator agent against a discriminator agent that collaborate to iteratively refine the Rust translation. On each iteration, the translator agent synthesizes and refines a Rust translation to pass an existing suite of tests, and then the discriminator agent finds new failing tests by constructing and refining a differential fuzzer over the C and Rust binaries. Across 63 real-world command-line C utilities, with an average size of 473 lines of code and the longest reaching thousands of lines in size, ACToR achieves over 90% test pass rate with zero human intervention. The improvement holds across seven agent-LLM configurations on our micro-benchmark, indicating that the loop is largely independent of the choice of underlying translator and LLM. Compared to a non-adversarial, coverage-driven test-generation baseline, ACToR improves correctness by up to 36.7%. When applied on top of one recent translator, C2SaferRust, ACToR further improves the validation pass rate by 16.6%.

cs.SE↗

arXiv2025

Securing Mixed Rust with Hardware Capabilities

The Rust programming language enforces three basic Rust principles, namely ownership, borrowing, and AXM (Aliasing Xor Mutability) to prevent security bugs such as memory safety violations and data races. However, Rust projects often have mixed code, i.e., code that also uses unsafe Rust, FFI (Foreign Function Interfaces), and inline assembly for low-level control. The Rust compiler is unable to statically enforce Rust principles in mixed Rust code which can lead to many security vulnerabilities. In this paper, we propose CapsLock, a security enforcement mechanism that can run at the level of machine code and detect Rust principle violations at run-time in mixed code. CapsLock is kept simple enough to be implemented into recent capability-based hardware abstractions that provide low-cost spatial memory safety. CapsLock introduces a novel revoke-on-use abstraction for capability-based designs, wherein accessing a memory object via a capability implicitly invalidates certain other capabilities pointing to it, thereby also providing temporal memory safety automatically, without requiring software to explicitly specify such invalidation. Thus, CapsLock is the first mechanism capable of providing cross-language enforcement of Rust principles. We implemented a prototype of CapsLock on QEMU. Evaluation results show that CapsLock is highly compatible with existing Rust code (passing 99.7% of the built-in test cases of the 100 most popular crates) and flags Rust principle violations in real-world Rust projects that use FFI or inline assembly. We discovered 8 previously unknown bugs in such crates in our experiments.

cs.CR↗

arXiv2025

A Practical and Secure Byzantine Robust Aggregator

In machine learning security, one is often faced with the problem of removing outliers from a given set of high-dimensional vectors when computing their average. For example, many variants of data poisoning attacks produce gradient vectors during training that are outliers in the distribution of clean gradients, which bias the computed average used to derive the ML model. Filtering them out before averaging serves as a generic defense strategy. Byzantine robust aggregation is an algorithmic primitive which computes a robust average of vectors, in the presence of an $\epsilon$ fraction of vectors which may have been arbitrarily and adaptively corrupted, such that the resulting bias in the final average is provably bounded. In this paper, we give the first robust aggregator that runs in quasi-linear time in the size of input vectors and provably has near-optimal bias bounds. Our algorithm also does not assume any knowledge of the distribution of clean vectors, nor does it require pre-computing any filtering thresholds from it. This makes it practical to use directly in standard neural network training procedures. We empirically confirm its expected runtime efficiency and its effectiveness in nullifying 10 different ML poisoning attacks.

cs.CR↗

arXiv2025

Program Skeletons for Automated Program Translation

Translating software between programming languages is a challenging task, for which automated techniques have been elusive and hard to scale up to larger programs. A key difficulty in cross-language translation is that one has to re-express the intended behavior of the source program into idiomatic constructs of a different target language. This task needs abstracting away from the source language-specific details, while keeping the overall functionality the same. In this work, we propose a novel and systematic approach for making such translation amenable to automation based on a framework we call program skeletons. A program skeleton retains the high-level structure of the source program by abstracting away and effectively summarizing lower-level concrete code fragments, which can be mechanically translated to the target programming language. A skeleton, by design, permits many different ways of filling in the concrete implementation for fragments, which can work in conjunction with existing data-driven code synthesizers. Most importantly, skeletons can conceptually enable sound decomposition, i.e., if each individual fragment is correctly translated, taken together with the mechanically translated skeleton, the final translated program is deemed to be correct as a whole. We present a prototype system called Skel embodying the idea of skeleton-based translation from Python to JavaScript. Our results show promising scalability compared to prior works. For 9 real-world Python programs, some with more than about 1k lines of code, 95% of their code fragments can be automatically translated, while about 5% require manual effort. All the final translations are correct with respect to whole-program test suites.

cs.PL↗

arXiv2025

Anvil: A General-Purpose Timing-Safe Hardware Description Language

Expressing hardware designs using hardware description languages (HDLs) routinely involves using stateless signals whose values change according to their underlying registers. Unintended behaviours can arise when the stored values in these underlying registers are mutated while their dependent signals are expected to remain constant across multiple cycles. Such timing hazards are common because, with a few exceptions, existing HDLs lack abstractions for values that remain unchanged over multiple clock cycles, delegating this responsibility to hardware designers. Designers must then carefully decide whether a value should remain unchanged, sometimes even across hardware modules. This paper proposes Anvil, an HDL which statically prevents timing hazards with a novel type system. Anvil is the only HDL we know of that guarantees timing safety, i.e., absence of timing hazards, without sacrificing expressiveness for cycle-level timing control or dynamic timing behaviours. Unlike many HLS languages that abstract away the differences between registers and signals, Anvil's type system exposes them fully while capturing the timing relationships between register value mutations and signal usages to enforce timing safety. This, in turn, enables safe composition of communicating hardware modules by static enforcement of timing contracts that encode timing constraints on shared signals. Such timing contracts can be specified parametric on abstract time points that can vary during run-time, allowing the type system to statically express dynamic timing behaviour. We have implemented Anvil and successfully used it to implement key timing-sensitive modules, comparing them against open-source SystemVerilog counterparts to demonstrate the practicality and expressiveness of the generated hardware.

cs.AR↗

arXiv2025

Model Provenance Testing for Large Language Models

Large language models are increasingly customized through fine-tuning and other adaptations, creating challenges in enforcing licensing terms and managing downstream impacts. Tracking model origins is crucial both for protecting intellectual property and for identifying derived models when biases or vulnerabilities are discovered in foundation models. We address this challenge by developing a framework for testing model provenance: Whether one model is derived from another. Our approach is based on the key observation that real-world model derivations preserve significant similarities in model outputs that can be detected through statistical analysis. Using only black-box access to models, we employ multiple hypothesis testing to compare model similarities against a baseline established by unrelated models. On two comprehensive real-world benchmarks spanning models from 30M to 4B parameters and comprising over 600 models, our tester achieves 90-95% precision and 80-90% recall in identifying derived models. These results demonstrate the viability of systematic provenance verification in production environments even when only API access is available.

cs.CR↗

arXiv2024

Translating C To Rust: Lessons from a User Study

Rust aims to offer full memory safety for programs, a guarantee that untamed C programs do not enjoy. How difficult is it to translate existing C code to Rust? To get a complementary view from that of automatic C to Rust translators, we report on a user study asking humans to translate real-world C programs to Rust. Our participants are able to produce safe Rust translations, whereas state-of-the-art automatic tools are not able to do so. Our analysis highlights that the high-level strategy taken by users departs significantly from those of automatic tools we study. We also find that users often choose zero-cost (static) abstractions for temporal safety, which addresses a predominant component of runtime costs in other full memory safety defenses. User-provided translations showcase a rich landscape of specialized strategies to translate the same C program in different ways to safe Rust, which future automatic translators can consider.

cs.SE↗

arXiv2024

CLUE-MARK: Watermarking Diffusion Models using CLWE

As AI-generated images become widespread, reliable watermarking is essential for content verification, copyright enforcement, and combating disinformation. Existing techniques rely on heuristic approaches and lack formal guarantees of undetectability, making them vulnerable to steganographic attacks that can expose or erase the watermark. Additionally, these techniques often degrade output quality by introducing perceptible changes, which is not only undesirable but an important barrier to adoption in practice. In this work, we introduce CLUE-Mark, the first provably undetectable watermarking scheme for diffusion models. CLUE-Mark requires no changes to the model being watermarked, is computationally efficient, and because it is provably undetectable is guaranteed to have no impact on model output quality. Our approach leverages the Continuous Learning With Errors (CLWE) problem -- a cryptographically hard lattice problem -- to embed watermarks in the latent noise vectors used by diffusion models. By proving undetectability via reduction from a cryptographically hard problem we ensure not only that the watermark is imperceptible to human observers or adhoc heuristics, but to \emph{any} efficient detector that does not have the secret key. CLUE-Mark allows multiple keys to be embedded, enabling traceability of images to specific users without altering model parameters. Empirical evaluations on state-of-the-art diffusion models confirm that CLUE-Mark achieves high recoverability, preserves image quality, and is robust to minor perturbations such JPEG compression and brightness adjustments. Uniquely, CLUE-Mark cannot be detected nor removed by recent steganographic attacks.

cs.CR↗

arXiv2024

Scalable Neural Network Training over Distributed Graphs

Graph neural networks (GNNs) fuel diverse machine learning tasks involving graph-structured data, ranging from predicting protein structures to serving personalized recommendations. Real-world graph data must often be stored distributed across many machines not just because of capacity constraints, but because of compliance with data residency or privacy laws. In such setups, network communication is costly and becomes the main bottleneck to train GNNs. Optimizations for distributed GNN training have targeted data-level improvements so far -- via caching, network-aware partitioning, and sub-sampling -- that work for data center-like setups where graph data is accessible to a single entity and data transfer costs are ignored. We present RETEXO, the first framework which eliminates the severe communication bottleneck in distributed GNN training while respecting any given data partitioning configuration. The key is a new training procedure, lazy message passing, that reorders the sequence of training GNN elements. RETEXO achieves 1-2 orders of magnitude reduction in network data costs compared to standard GNN training, while retaining accuracy. RETEXO scales gracefully with increasing decentralization and decreasing bandwidth. It is the first framework that can be used to train GNNs at all network decentralization levels -- including centralized data-center networks, wide area networks, proximity networks, and edge networks.

cs.LG↗

arXiv2023

Attacking Byzantine Robust Aggregation in High Dimensions

Training modern neural networks or models typically requires averaging over a sample of high-dimensional vectors. Poisoning attacks can skew or bias the average vectors used to train the model, forcing the model to learn specific patterns or avoid learning anything useful. Byzantine robust aggregation is a principled algorithmic defense against such biasing. Robust aggregators can bound the maximum bias in computing centrality statistics, such as mean, even when some fraction of inputs are arbitrarily corrupted. Designing such aggregators is challenging when dealing with high dimensions. However, the first polynomial-time algorithms with strong theoretical bounds on the bias have recently been proposed. Their bounds are independent of the number of dimensions, promising a conceptual limit on the power of poisoning attacks in their ongoing arms race against defenses. In this paper, we show a new attack called HIDRA on practical realization of strong defenses which subverts their claim of dimension-independent bias. HIDRA highlights a novel computational bottleneck that has not been a concern of prior information-theoretic analysis. Our experimental evaluation shows that our attacks almost completely destroy the model performance, whereas existing attacks with the same goal fail to have much effect. Our findings leave the arms race between poisoning attacks and provable defenses wide open.

cs.CR↗

arXiv2023

Capstone: A Capability-based Foundation for Trustless Secure Memory Access (Extended Version)

Capability-based memory isolation is a promising new architectural primitive. Software can access low-level memory only via capability handles rather than raw pointers, which provides a natural interface to enforce security restrictions. Existing architectural capability designs such as CHERI provide spatial safety, but fail to extend to other memory models that security-sensitive software designs may desire. In this paper, we propose Capstone, a more expressive architectural capability design that supports multiple existing memory isolation models in a trustless setup, i.e., without relying on trusted software components. We show how Capstone is well-suited for environments where privilege boundaries are fluid (dynamically extensible), memory sharing/delegation are desired both temporally and spatially, and where such needs are to be balanced with availability concerns. Capstone can also be implemented efficiently. We present an implementation sketch and through evaluation show that its overhead is below 50% in common use cases. We also prototype a functional emulator for Capstone and use it to demonstrate the runnable implementations of six real-world memory models without trusted software components: three types of enclave-based TEEs, a thread scheduler, a memory allocator, and Rust-style memory safety -- all within the interface of Capstone.

cs.CR↗

arXiv2023

User-Customizable Transpilation of Scripting Languages

A transpiler converts code from one programming language to another. Many practical uses of transpilers require the user to be able to guide or customize the program produced from a given input program. This customizability is important for satisfying many application-specific goals for the produced code such as ensuring performance, readability, maintainability, compatibility, and so on. Conventional transpilers are deterministic rule-driven systems often written without offering customizability per user and per program. Recent advances in transpilers based on neural networks offer some customizability to users, e.g. through interactive prompts, but they are still difficult to precisely control the production of a desired output. Both conventional and neural transpilation also suffer from the "last mile" problem: they produce correct code on average, i.e., on most parts of a given program, but not necessarily for all parts of it. We propose a new transpilation approach that offers fine-grained customizability and reusability of transpilation rules created by others, without burdening the user to understand the global semantics of the given source program. Our approach is mostly automatic and incremental, i.e., constructs translation rules needed to transpile the given program as per the user's guidance piece-by-piece. We implement the transpiler as a tool called DuoGlot, which translates Python to Javascript programs, and evaluate it on the popular GeeksForGeeks benchmarks. DuoGlot achieves 90% translation accuracy and so it outperforms all existing translators, while it produces readable code. We evaluate DuoGlot on two additional benchmarks, containing more challenging and longer programs, and similarly observe improved accuracy.

cs.PL↗

arXiv2022

Membership Inference Attacks and Generalization: A Causal Perspective

Membership inference (MI) attacks highlight a privacy weakness in present stochastic training methods for neural networks. It is not well understood, however, why they arise. Are they a natural consequence of imperfect generalization only? Which underlying causes should we address during training to mitigate these attacks? Towards answering such questions, we propose the first approach to explain MI attacks and their connection to generalization based on principled causal reasoning. We offer causal graphs that quantitatively explain the observed MI attack performance achieved for $6$ attack variants. We refute several prior non-quantitative hypotheses that over-simplify or over-estimate the influence of underlying causes, thereby failing to capture the complex interplay between several factors. Our causal models also show a new connection between generalization and MI attacks via their shared causal factors. Our causal models have high predictive power ($0.90$), i.e., their analytical predictions match with observations in unseen experiments often, which makes analysis via them a pragmatic alternative.

cs.LG↗

arXiv2022

LPGNet: Link Private Graph Networks for Node Classification

Classification tasks on labeled graph-structured data have many important applications ranging from social recommendation to financial modeling. Deep neural networks are increasingly being used for node classification on graphs, wherein nodes with similar features have to be given the same label. Graph convolutional networks (GCNs) are one such widely studied neural network architecture that perform well on this task. However, powerful link-stealing attacks on GCNs have recently shown that even with black-box access to the trained model, inferring which links (or edges) are present in the training graph is practical. In this paper, we present a new neural network architecture called LPGNet for training on graphs with privacy-sensitive edges. LPGNet provides differential privacy (DP) guarantees for edges using a novel design for how graph edge structure is used during training. We empirically show that LPGNet models often lie in the sweet spot between providing privacy and utility: They can offer better utility than "trivially" private architectures which use no edge information (e.g., vanilla MLPs) and better resilience against existing link-stealing attacks than vanilla GCNs which use the full edge structure. LPGNet also offers consistently better privacy-utility tradeoffs than DPGCN, which is the state-of-the-art mechanism for retrofitting differential privacy into conventional GCNs, in most of our evaluated datasets.

cs.LG↗

arXiv2021

Using Throughput-Centric Byzantine Broadcast to Tolerate Malicious Majority in Blockchains

Fault tolerance of a blockchain is often characterized by the fraction $f$ of "adversarial power" that it can tolerate in the system. Despite the fast progress in blockchain designs in recent years, existing blockchain systems can still only tolerate $f$ below $0.5$. Can practically usable blockchains tolerate a malicious majority, i.e., $f$ above $0.5$? This work presents a positive answer to this question. We first note that the well-known impossibility of {\em byzantine consensus} for $f$ above $0.5$ does not carry over to blockchains. To tolerate $f$ above $0.5$, we use {\em byzantine broadcast}, instead of byzantine consensus, as the core of the blockchain. A major obstacle in doing so, however, is that the resulting blockchain may have extremely low throughput. To overcome this central technical challenge, we propose a novel byzantine broadcast protocol OverlayBB, that can tolerate $f$ above $0.5$ while achieving good throughput. Using OverlayBB as the core, we present the design, implementation, and evaluation of a novel Proof-of-Stake blockchain called BCube. BCube can tolerate a malicious majority, while achieving practically usable transaction throughput and confirmation latency in our experiments with $10000$ nodes and under $f = 0.7$. To our knowledge, BCube is the first blockchain that can achieve such properties.

cs.CR↗