Searcharxiv⌕ Search

arXiv subjects

Ahmad-Reza Sadeghi

Publications and source records attributed to Ahmad-Reza Sadeghi.

At least 19 recordsLinked to original sources
arXiv2026

JevVibe: Efficient Classification-Guided Secure Code Generation

Large language models can generate functionally correct code that still contains security weaknesses, motivating repair pipelines that first diagnose a weakness type before deciding how to fix it. The Common Weakness Enumeration (CWE) provides a standardized vocabulary for such diagnoses, but asking an autoregressive language model to generate a CWE label and extracting it from the response raises questions about output validity, speed, and cost, as well as accuracy. We evaluate Jev, a decision model that instead selects directly from a declared set of candidates and returns a probability for each, against six open-weight autoregressive models and a frontier proprietary model, GPT-5.6-Sol, on a controlled 50-way CWE classification task over 1,916 CyberSecEval benchmark examples. Jev outperforms all six open-weight baselines on every classification and ranking metric, while its comparison with GPT-5.6-Sol depends on the metric: GPT-5.6-Sol achieves higher Top-1 accuracy and Macro-F1, whereas Jev achieves higher Top-3 and Top-5 accuracy and a nearly identical MRR, at $6.27\times$ lower median API latency and $55.9\times$ lower estimated API cost. We further build JevVibe, a diagnosis-guided repair agent that uses predicted CWE labels to repair code generated by Qwen2.5-Coder-32B-Instruct. With Jev providing the diagnosis, the agent increases the detector-measured security pass rate from 63.5% before repair to 70.7%, compared with 66.1% for LLM-guided repair. These results show that JevVibe is effective at improving the security of generated code, with Jev providing reliable and efficient CWE classification.

cs.CR↗
arXiv2026

From Verdict to Diagnosis: Attributable Security Review of Pull Requests

Automated code reviewers are increasingly used as gates on pull requests (PRs), yet evaluations measure whether they block a malicious change. A block may be triggered by an unrelated issue rather than the vulnerability that makes the PR unsafe; fixing the reported issue can leave the target defect exploitable. We call this discrepancy the Verdict-Diagnosis (VD) gap. We present MalPR-Bench, a mechanism-grounded benchmark of 89 malicious PRs and 50 paired benign controls across 44 repositories and eight language families. Each malicious case has a pre-committed rubric specifying the target vulnerability, accepted mechanism descriptions, required repository evidence, and off-target findings receiving no credit. Reviews are scored separately for verdict correctness, target-vulnerability identification, and evidence validation; an attributable block requires all three. We introduce PRGuard, an attributable PR security reviewer that constructs candidate vulnerabilities and validates their premises against repository evidence using deterministic, non-executing tools and bounded retrieval. Across 31 common-coverage held-out malicious PRs, PRGuard and CodeRabbit produce similar blocking totals (22/31 vs. 24/31), but PRGuard identifies 22 target vulnerabilities versus 16 for CodeRabbit, a 1.38x difference. On 14 absence-type cases, both block 9, while PRGuard identifies 9 targets versus 3. CodeRabbit identifies 16/24 targets when required evidence lies within touched files and 0/7 when validation requires evidence outside them. Finally, PRGuard uncovers twelve previously undisclosed, proof-of-concept-backed vulnerabilities across five projects. PRGuard/DeepSeek and CodeRabbit both block 10/12 discovery PRs, but produce 10/12 and 4/12 attributable blocks, respectively. Thus, verdict-only evaluation can substantially overstate the security value of automated review.

cs.CR↗
arXiv2026

Walma: Learning to See Memory Corruption in WebAssembly

WebAssembly's (Wasm) monolithic linear memory turns a single memory-corruption bug into a bidirectional threat: a compromised module can attack its embedding host, and a malicious host can tamper with a trusted module's state. Existing defenses require custom runtimes or source changes, and none can verify runtime integrity under an adversarial host. We present Walma, a framework for WebAssembly linear memory attestation that renders snapshots of linear memory as grayscale images and classifies them with a convolutional neural network. The network reads the row-aligned structure that compiled code imposes on memory and detects corruption that byte and texture statistics miss. On structured targets this extends to tampering that no program input triggers, such as direct memory writes by a malicious host. On real-world CVE-affected applications, hiding corruption from Walma's verdict requires overwriting a large fraction of memory, hundreds of kilobytes to megabytes on our targets. A Shannon-entropy analysis of benign memory bounds the class of out-of-band edits that the learned prior can detect. Across the 53 binaries of the WABench suite, attestation costs 1.07x--1.69x at the host boundary, making continuous, state-centric memory attestation for Wasm practical.

cs.CR↗
arXiv2026

AttackonCTF: Defending Hardware Security Competition Benchmarks in the Age of LLMs

Hardware security competitions such as HackTheSilicon serve as benchmarking platforms for evaluating vulnerability detection methods and for training humans and AI. However, our study reveals that LLMs threaten their validity. Instead of genuine security reasoning, detectors exploit a diff-style syntactic comparison, achieving an 83% detection rate, undermining fair evaluation. To mitigate this, we propose the first LLM-oriented, semantics-preserving obfuscation framework for these benchmarks. Unlike IP-protection approaches, it applies human-readable transformations and controlled diff-noise while preserving functionality. On HackTheSilicon, the framework reduces LLM-based detection accuracy by 50% with only 10% obfuscation and by 78.6% under complete obfuscation, restoring benchmark reliability.

cs.CR↗
arXiv2026

NeST: Neuron Selective Tuning for LLM Safety

Safety alignment is essential for the responsible deployment of Large Language Models (LLMs). Yet, existing approaches often rely on heavyweight fine-tuning that is costly to update, audit, and maintain across model families. Full fine-tuning incurs substantial computational and storage overhead, while parameter-efficient methods, e.g., Low-Rank Adaptation (LoRA), trade efficiency for inconsistent safety gains and sensitivity to design choices. Safety intervention mechanisms reduce unsafe outputs without modifying model weights, but do not directly shape or preserve the internal representations that govern safety behavior. We present NeST, a Neuron-Selective Tuning framework for efficient post-hoc safety alignment. NeST identifies safety-relevant feed-forward neurons via activation probing on vanilla harmful and benign prompts, clusters neurons with similar activation profiles, and trains shared cluster-level updates while freezing the rest of the model. Importantly, NeST is trained only on vanilla malicious prompts, without using jailbreak-specific attack data, yet generalizes robustly to diverse jailbreaks. The learned updates are then folded into the original weights, incurring no inference-time overhead. Evaluated on 14 open-weight language and multimodal models, NeST outperforms lightweight baselines and approaches full fine-tuning robustness with significantly fewer trainable parameters. On text-only models, NeST reduces average jailbreak attack success rate from 44.5% to 1.1% while training only 0.4M parameters on average. Across multimodal settings, it reduces ASR from 55.3% to 1.1%, and for downstream fine-tuned variants, it restores safety by reducing ASR from 53.8% to 0.8%. These results show that robust, maintainable safety alignment can be achieved by concentrating adaptation on localized, functionally coherent safety structures.

cs.CR↗
arXiv2026

GoodVibe: Security-by-Vibe for LLM-Based Code Generation

Large language models (LLMs) are increasingly used for code generation in fast, informal development workflows, often referred to as vibe coding, where speed and convenience are prioritized, and security requirements are rarely made explicit. In this setting, models frequently produce functionally correct but insecure code, creating a growing security risk. Existing approaches to improving code security rely on full-parameter fine-tuning or parameter-efficient adaptations, which are either costly and prone to catastrophic forgetting or operate at coarse granularity with limited interpretability and control. We present GoodVibe, a neuron-level framework for improving the security of code language models by default. GoodVibe is based on the key insight that security-relevant reasoning is localized to a small subset of neurons. We identify these neurons using gradient-based attribution from a supervised security task and perform neuron-selective fine-tuning that updates only this security-critical subspace. To further reduce training cost, we introduce activation-driven neuron clustering, enabling structured updates with minimal overhead. We evaluate GoodVibe on six LLMs across security-critical programming languages, including C++, Java, Swift, and Go. GoodVibe substantially improves the security of generated code while preserving general model utility, achieving up to a 2.5x improvement over base models, achieving performance competitive with full fine-tuning while using over 4,700x fewer trainable parameters, and reducing training computation by more than 3.6x compared to the parameter-efficient baseline (LoRA). Our results demonstrate that neuron-level optimization offers an effective and scalable approach to securing code generation without sacrificing generality.

cs.CR↗
arXiv2026

Measuring Security Without Fooling Ourselves: Why Benchmarking Agents Is Hard

The benchmarks used to evaluate AI agents in security-critical roles suffer from crucial weaknesses. Building on recent empirical evidence, we characterize three core challenges that undermine security evaluations: benchmark vulnerabilities, temporal staleness, and runtime uncertainty. We then outline practical directions toward building more robust and trustworthy evaluation frameworks.

cs.CR↗
arXiv2026

LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections

Split learning (SL) enables collaborative training by partitioning a neural network across clients and a central server, but the cut-layer interface introduces a key challenge: high-dimensional activations incur substantial communication overhead while exposing representations vulnerable to reconstruction attacks. Existing approaches typically address efficiency or privacy in isolation, relying on additional mechanisms such as sparsification, quantization, or noise injection. We propose LightSplit, which limits information exposure and reduces communication overhead by applying a lightweight fixed orthogonal random projection at the cut layer. Based on Shannon's information theory, this projection acts as an information bottleneck that restricts instance-specific information and suppresses exploitable per-sample signals. By transmitting low-dimensional projections instead of raw activations, the server operates on lifted representations without requiring architectural modifications, ensuring compatibility with existing SL architectures. By avoiding additional trainable components on the client, the method remains lightweight and suitable for edge devices while preserving end-to-end differentiability via exact gradient propagation. As the projection is non-invertible, part of the original representation is irreversibly discarded at the client, LightSplit reduces the information available for reconstruction and limits information exposure. We extensively evaluate LightSplit on state-of-the-art benchmarks in both IID and non-IID settings across varying projection dimensions and client scales. Our results show that the method retains more than 95% of the baseline accuracy at up to 32x reduction in transmitted dimensionality while maintaining stable training dynamics.

cs.LG↗
arXiv2026

Towards Remote Attestation of Microarchitectural Attacks: The Case of Rowhammer

Microarchitectural vulnerabilities increasingly undermine the assumption that hardware can be treated as a reliable root of trust. Prevention mechanisms often lag behind evolving attack techniques, leaving deployed systems unable to assume continued trustworthiness. We propose a shift from prevention to detection through microarchitectural-aware remote attestation. As a first instantiation of this idea, we present HammerWatch, a Rowhammer-aware remote attestation protocol that enables an external verifier to assess whether a system exhibits hardware-induced disturbance behavior. HammerWatch leverages memory-level evidence available on commodity platforms, specifically Machine-Check Exceptions (MCEs) from ECC DRAM and counter-based indicators from Per-Row Activation Counting (PRAC), and protects these measurements against kernel-level adversaries using TPM-anchored hash chains. We implement HammerWatch on commodity hardware and evaluate it on 20000 simulated benign and malicious access patterns. Our results show that the verifier reliably distinguishes Rowhammer-like behavior from benign operation under conservative heuristics, demonstrating that detection-oriented attestation is feasible and can complement incomplete prevention mechanisms

cs.CR↗
arXiv2026

NeuroStrike: Neuron-Level Attacks on Aligned LLMs

Safety alignment is critical for the ethical deployment of large language models (LLMs), guiding them to avoid generating harmful or unethical content. Current alignment techniques, such as supervised fine-tuning and reinforcement learning from human feedback, remain fragile and can be bypassed by carefully crafted adversarial prompts. Unfortunately, such attacks rely on trial and error, lack generalizability across models, and are constrained by scalability and reliability. This paper presents NeuroStrike, a novel and generalizable attack framework that exploits a fundamental vulnerability introduced by alignment techniques: the reliance on sparse, specialized safety neurons responsible for detecting and suppressing harmful inputs. We apply NeuroStrike to both white-box and black-box settings: In the white-box setting, NeuroStrike identifies safety neurons through feedforward activation analysis and prunes them during inference to disable safety mechanisms. In the black-box setting, we propose the first LLM profiling attack, which leverages safety neuron transferability by training adversarial prompt generators on open-weight surrogate models and then deploying them against black-box and proprietary targets. We evaluate NeuroStrike on over 20 open-weight LLMs from major LLM developers. By removing less than 0.6% of neurons in targeted layers, NeuroStrike achieves an average attack success rate (ASR) of 76.9% using only vanilla malicious prompts. Moreover, Neurostrike generalizes to four multimodal LLMs with 100% ASR on unsafe image inputs. Safety neurons transfer effectively across architectures, raising ASR to 78.5% on 11 fine-tuned models and 77.7% on five distilled models. The black-box LLM profiling attack achieves an average ASR of 63.7% across five black-box models, including the Google Gemini family.

cs.CR↗
arXiv2026

AegisSat: Securing AI-Enabled SoC FPGA Satellite Platforms

The increasing adoption of System-on-Chip Field-Programmable Gate Arrays (SoC FPGAs) in AI-enabled satellite systems, valued for their reconfigurability and in-orbit update capabilities, introduces significant security challenges. Compromised updates can lead to performance degradation, service disruptions, or adversarial manipulation of mission outcomes. To address these risks, this paper proposes a comprehensive security framework, AegisSat. It ensures the integrity and resilience of satellite platforms by (i) integrating cryptographically-based secure boot mechanisms to establish a trusted computing base; (ii) enforcing strict runtime resource isolation; (iii) employing authenticated procedures for in-orbit reconfiguration and AI model updates to prevent unauthorized modifications; and (iv) providing robust rollback capabilities to recover from boot and update failures and maintain system stability. To further support our claims, we conducted experiments demonstrating the integration of these mechanisms on contemporary SoC FPGA devices. This defense-in-depth framework is crucial for space applications, where physical access is impossible and systems must operate reliably over extended periods, thereby enhancing the trustworthiness of SoC FPGA-based satellite systems and enabling secure and resilient AI operations in orbit.

cs.CR↗
arXiv2025

Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer

Modern CPUs are black boxes, proprietary, and increasingly characterized by sophisticated microarchitectural flaws that evade traditional analysis. While some of these critical vulnerabilities have been uncovered through cumbersome manual effort, building an automated and systematic vulnerability detection framework for real-world post-silicon processors remains a challenge. In this paper, we present Fuzzilicon, the first post-silicon fuzzing framework for real-world x86 CPUs that brings deep introspection into the microcode and microarchitectural layers. Fuzzilicon automates the discovery of vulnerabilities that were previously only detectable through extensive manual reverse engineering, and bridges the visibility gap by introducing microcode-level instrumentation. At the core of Fuzzilicon is a novel technique for extracting feedback directly from the processor's microarchitecture, enabled by reverse-engineering Intel's proprietary microcode update interface. We develop a minimally intrusive instrumentation method and integrate it with a hypervisor-based fuzzing harness to enable precise, feedback-guided input generation, without access to Register Transfer Level (RTL). Applied to Intel's Goldmont microarchitecture, Fuzzilicon introduces 5 significant findings, including two previously unknown microcode-level speculative-execution vulnerabilities. Besides, the Fuzzilicon framework automatically rediscover the $μ$Spectre class of vulnerabilities, which were detected manually in the previous work. Fuzzilicon reduces coverage collection overhead by up to 31$\times$ compared to baseline techniques and achieves 16.27% unique microcode coverage of hookable locations, the first empirical baseline of its kind. As a practical, coverage-guided, and scalable approach to post-silicon fuzzing, Fuzzilicon establishes a new foundation to automate the discovery of complex CPU vulnerabilities.

cs.CR↗
arXiv2025

GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMs

Mixture-of-Experts (MoE) architectures have advanced the scaling of Large Language Models (LLMs) by activating only a sparse subset of parameters per input, enabling state-of-the-art performance with reduced computational cost. As these models are increasingly deployed in critical domains, understanding and strengthening their alignment mechanisms is essential to prevent harmful outputs. However, existing LLM safety research has focused almost exclusively on dense architectures, leaving the unique safety properties of MoEs largely unexamined. The modular, sparsely-activated design of MoEs suggests that safety mechanisms may operate differently than in dense models, raising questions about their robustness. In this paper, we present GateBreaker, the first training-free, lightweight, and architecture-agnostic attack framework that compromises the safety alignment of modern MoE LLMs at inference time. GateBreaker operates in three stages: (i) gate-level profiling, which identifies safety experts disproportionately routed on harmful inputs, (ii) expert-level localization, which localizes the safety structure within safety experts, and (iii) targeted safety removal, which disables the identified safety structure to compromise the safety alignment. Our study shows that MoE safety concentrates within a small subset of neurons coordinated by sparse routing. Selective disabling of these neurons, approximately 3% of neurons in the targeted expert layers, significantly increases the averaged attack success rate (ASR) from 7.4% to 64.9% against the eight latest aligned MoE LLMs with limited utility degradation. These safety neurons transfer across models within the same family, raising ASR from 17.9% to 67.7% with one-shot transfer attack. Furthermore, GateBreaker generalizes to five MoE vision language models (VLMs) with 60.9% ASR on unsafe image inputs.

cs.CR↗
arXiv2025

GoldenFuzz: Generative Golden Reference Hardware Fuzzing

Modern hardware systems, driven by demands for high performance and application-specific functionality, have grown increasingly complex, introducing large surfaces for bugs and security-critical vulnerabilities. Fuzzing has emerged as a scalable solution for discovering such flaws. Yet, existing hardware fuzzers suffer from limited semantic awareness, inefficient test refinement, and high computational overhead due to reliance on slow device simulation. In this paper, we present GoldenFuzz, a novel two-stage hardware fuzzing framework that partially decouples test case refinement from coverage and vulnerability exploration. GoldenFuzz leverages a fast, ISA-compliant Golden Reference Model (GRM) as a ``digital twin'' of the Device Under Test (DUT). It fuzzes the GRM first, enabling rapid, low-cost test case refinement, accelerating deep architectural exploration and vulnerability discovery on DUT. During the fuzzing pipeline, GoldenFuzz iteratively constructs test cases by concatenating carefully chosen instruction blocks that balance the subtle inter- and intra-instructions quality. A feedback-driven mechanism leveraging insights from both high- and low-coverage samples further enhances GoldenFuzz's capability in hardware state exploration. Our evaluation of three RISC-V processors, RocketChip, BOOM, and CVA6, demonstrates that GoldenFuzz significantly outperforms existing fuzzers in achieving the highest coverage with minimal test case length and computational overhead. GoldenFuzz uncovers all known vulnerabilities and discovers five new ones, four of which are classified as highly severe with CVSS v3 severity scores exceeding seven out of ten. It also identifies two previously unknown vulnerabilities in the commercial BA51-H core extension.

cs.CR↗
arXiv2025

ReFuzz: Reusing Tests for Processor Fuzzing with Contextual Bandits

Processor designs rely on iterative modifications and reuse well-established designs. However, this reuse of prior designs also leads to similar vulnerabilities across multiple processors. As processors grow increasingly complex with iterative modifications, efficiently detecting vulnerabilities from modern processors is critical. Inspired by software fuzzing, hardware fuzzing has recently demonstrated its effectiveness in detecting processor vulnerabilities. Yet, to our best knowledge, existing processor fuzzers fuzz each design individually, lacking the capability to understand known vulnerabilities in prior processors to fine-tune fuzzing to identify similar or new variants of vulnerabilities. To address this gap, we present ReFuzz, an adaptive fuzzing framework that leverages contextual bandit to reuse highly effective tests from prior processors to fuzz a processor-under-test (PUT) within a given ISA. By intelligently mutating tests that trigger vulnerabilities in prior processors, ReFuzz effectively detects similar and new variants of vulnerabilities in PUTs. ReFuzz uncovered three new security vulnerabilities and two new functional bugs. ReFuzz detected one vulnerability by reusing a test that triggers a known vulnerability in a prior processor. One functional bug exists across three processors that share design modules. The second bug has two variants. Additionally, ReFuzz reuses highly effective tests to enhance efficiency in coverage, achieving an average 511.23x coverage speedup and up to 9.33% more total coverage, compared to existing fuzzers.

cs.CR↗
arXiv2025

ZORRO: Zero-Knowledge Robustness and Privacy for Split Learning (Full Version)

Split Learning (SL) is a distributed learning approach that enables resource-constrained clients to collaboratively train deep neural networks (DNNs) by offloading most layers to a central server while keeping in- and output layers on the client-side. This setup enables SL to leverage server computation capacities without sharing data, making it highly effective in resource-constrained environments dealing with sensitive data. However, the distributed nature enables malicious clients to manipulate the training process. By sending poisoned intermediate gradients, they can inject backdoors into the shared DNN. Existing defenses are limited by often focusing on server-side protection and introducing additional overhead for the server. A significant challenge for client-side defenses is enforcing malicious clients to correctly execute the defense algorithm. We present ZORRO, a private, verifiable, and robust SL defense scheme. Through our novel design and application of interactive zero-knowledge proofs (ZKPs), clients prove their correct execution of a client-located defense algorithm, resulting in proofs of computational integrity attesting to the benign nature of locally trained DNN portions. Leveraging the frequency representation of model partitions enables ZORRO to conduct an in-depth inspection of the locally trained models in an untrusted environment, ensuring that each client forwards a benign checkpoint to its succeeding client. In our extensive evaluation, covering different model architectures as well as various attack strategies and data scenarios, we show ZORRO's effectiveness, as it reduces the attack success rate to less than 6\% while causing even for models storing \numprint{1000000} parameters on the client-side an overhead of less than 10 seconds.

cs.CR↗
arXiv2025

SafeSplit: A Novel Defense Against Client-Side Backdoor Attacks in Split Learning (Full Version)

Split Learning (SL) is a distributed deep learning approach enabling multiple clients and a server to collaboratively train and infer on a shared deep neural network (DNN) without requiring clients to share their private local data. The DNN is partitioned in SL, with most layers residing on the server and a few initial layers and inputs on the client side. This configuration allows resource-constrained clients to participate in training and inference. However, the distributed architecture exposes SL to backdoor attacks, where malicious clients can manipulate local datasets to alter the DNN's behavior. Existing defenses from other distributed frameworks like Federated Learning are not applicable, and there is a lack of effective backdoor defenses specifically designed for SL. We present SafeSplit, the first defense against client-side backdoor attacks in Split Learning (SL). SafeSplit enables the server to detect and filter out malicious client behavior by employing circular backward analysis after a client's training is completed, iteratively reverting to a trained checkpoint where the model under examination is found to be benign. It uses a two-fold analysis to identify client-induced changes and detect poisoned models. First, a static analysis in the frequency domain measures the differences in the layer's parameters at the server. Second, a dynamic analysis introduces a novel rotational distance metric that assesses the orientation shifts of the server's layer parameters during training. Our comprehensive evaluation across various data distributions, client counts, and attack scenarios demonstrates the high efficacy of this dual analysis in mitigating backdoor attacks while preserving model utility.

cs.CR↗
arXiv2024

Fuzzerfly Effect: Hardware Fuzzing for Memory Safety

Hardware-level memory vulnerabilities severely threaten computing systems. However, hardware patching is inefficient or difficult postfabrication. We investigate the effectiveness of hardware fuzzing in detecting hardware memory vulnerabilities and highlight challenges and potential future research directions to enhance hardware fuzzing for memory safety.

cs.CR↗