Your Compiler is Backdooring Your Model: Understanding and Exploiting Compilation Inconsistency Vulnerabilities in Deep Learning Compilers

• URL: http://arxiv.org/abs/2509.11173v3
• Date: Mon, 27 Oct 2025 03:04:50 GMT
• Title: Your Compiler is Backdooring Your Model: Understanding and Exploiting Compilation Inconsistency Vulnerabilities in Deep Learning Compilers
• Abstract summary: Deep learning (DL) compilers are core infrastructure in modern DL systems.<n>Can an official, unmodified compiler alter a model's semantics during compilation and introduce hidden backdoors?<n>We study both adversarial and natural settings. In the adversarial case, we craft benign models where triggers have no effect pre-compilation but become effective backdoors after compilation.
• Score: 28.679695741222318
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Deep learning (DL) compilers are core infrastructure in modern DL systems, offering flexibility and scalability beyond vendor-specific libraries. This work uncovers a fundamental vulnerability in their design: can an official, unmodified compiler alter a model's semantics during compilation and introduce hidden backdoors? We study both adversarial and natural settings. In the adversarial case, we craft benign models where triggers have no effect pre-compilation but become effective backdoors after compilation. Tested on six models, three commercial compilers, and two hardware platforms, our attack yields 100% success on triggered inputs while preserving normal accuracy and remaining undetected by state-of-the-art detectors. The attack generalizes across compilers, hardware, and floating-point settings. In the natural setting, we analyze the top 100 HuggingFace models (including one with 220M+ downloads) and find natural triggers in 31 models. This shows that compilers can introduce risks even without adversarial manipulation. Our results reveal an overlooked threat: unmodified DL compilers can silently alter model semantics. To our knowledge, this is the first work to expose inherent security risks in DL compiler design, opening a new direction for secure and trustworthy ML.

Related papers

• Finding Compiler-Platform Interaction Bugs in Deep Learning Pipelines via Cross-Layer Constraints [6.16235197336957]
  Testing deep learning (DL) compilers is challenging as correctness depends on implicit constraints embedded throughout the compilation stack.<n>Existing testing approaches largely take type constraints to restrict input model generation.<n>We propose a scalable, automated DL compiler testing framework for, in tandem, finding compiler-platform interaction bugs.
  arXiv  Detail & Related papers  (2026-06-16T19:15:17Z)
• Decaf: Improving Neural Decompilation with Automatic Feedback and Search [64.02615739798277]
  High-level syntax, identifiers, and custom data types are generally lost as the compiler translates human-readable code to low-level machine code.<n>Generative AI models are a natural fit for reconstructing high-level syntax, identifiers, and types.<n>We argue that compiler feedback can be used to dramatically improve the semantic correctness of neural decompiler outputs via search.
  arXiv  Detail & Related papers  (2026-05-12T04:21:26Z)
• CIBER: A Comprehensive Benchmark for Security Evaluation of Code Interpreter Agents [27.35968236632966]
  LLM-based code interpreter agents are increasingly deployed in critical situations.<n>Existing benchmarks fail to capture the security risks arising from dynamic code execution, tool interactions, and multi-turn context.<n>We introduce CIBER, an automated benchmark that combines dynamic attack generation, isolated secure sandboxing, and state-aware evaluation.
  arXiv  Detail & Related papers  (2026-02-23T06:41:41Z)
• Verifiable Provenance of Software Artifacts with Zero-Knowledge Compilation [5.939983212292006]
  We propose a novel approach to verifiable provenance based on compiling software with zero-knowledge virtual machines (zkVMs)<n>By executing a compiler within a zkVM, our system produces both the compiled output and a cryptographic proof attesting that the compilation was performed on the claimed source code with the claimed compiler.<n>Our results show that zk-compilation is applicable to real-world software and provides strong security guarantees.
  arXiv  Detail & Related papers  (2026-02-12T12:36:36Z)
• Assimilation Matters: Model-level Backdoor Detection in Vision-Language Pretrained Models [71.44858461725893]
  Given a model fine-tuned by an untrusted third party, determining whether the model has been injected with a backdoor is a critical and challenging problem.<n>Existing detection methods usually rely on prior knowledge of training dataset, backdoor triggers and targets.<n>We introduce Assimilation Matters in DETection (AMDET), a novel model-level detection framework that operates without any such prior knowledge.
  arXiv  Detail & Related papers  (2025-11-29T06:20:00Z)
• Towards Effective, Stealthy, and Persistent Backdoor Attacks Targeting Graph Foundation Models [62.87838888016534]
  Graph Foundation Models (GFMs) are pre-trained on diverse source domains and adapted to unseen targets.<n>Backdoor attacks against GFMs are non-trivial due to three key challenges.<n>We propose GFM-BA, a novel Backdoor Attack model against Graph Foundation Models.
  arXiv  Detail & Related papers  (2025-11-22T08:52:09Z)
• Context-Guided Decompilation: A Step Towards Re-executability [50.71992919223209]
  Binary decompilation plays an important role in software security analysis, reverse engineering and malware understanding.<n>Recent advances in large language models (LLMs) have enabled neural decompilation, but the generated code is typically only semantically plausible.<n>We propose ICL4Decomp, a hybrid decompilation framework that leverages in-context learning (ICL) to guide LLMs toward generating re-executable source code.
  arXiv  Detail & Related papers  (2025-11-03T17:21:39Z)
• Backdoor Unlearning by Linear Task Decomposition [69.91984435094157]
  Foundation models are highly susceptible to adversarial perturbations and targeted backdoor attacks.<n>Existing backdoor removal approaches rely on costly fine-tuning to override the harmful behavior.<n>This raises the question of whether backdoors can be removed without compromising the general capabilities of the models.
  arXiv  Detail & Related papers  (2025-10-16T16:18:07Z)
• Decompiling Smart Contracts with a Large Language Model [51.49197239479266]
  Despite Etherscan's 78,047,845 smart contracts deployed on (as of May 26, 2025), a mere 767,520 ( 1%) are open source.<n>This opacity necessitates the automated semantic analysis of on-chain smart contract bytecode.<n>We introduce a pioneering decompilation pipeline that transforms bytecode into human-readable and semantically faithful Solidity code.
  arXiv  Detail & Related papers  (2025-06-24T13:42:59Z)
• Robust Anti-Backdoor Instruction Tuning in LVLMs [53.766434746801366]
  We introduce a lightweight, certified-agnostic defense framework for large visual language models (LVLMs)<n>Our framework finetunes only adapter modules and text embedding layers under instruction tuning.<n>Experiments against seven attacks on Flickr30k and MSCOCO demonstrate that ours reduces their attack success rate to nearly zero.
  arXiv  Detail & Related papers  (2025-06-04T01:23:35Z)
• Exploiting the Vulnerability of Large Language Models via Defense-Aware Architectural Backdoor [0.24335447922683692]
  We introduce a new type of backdoor attack that conceals itself within the underlying model architecture. The add-on modules of model architecture layers can detect the presence of input trigger tokens and modify layer weights. We conduct extensive experiments to evaluate our attack methods using two model architecture settings on five different large language datasets.
  arXiv  Detail & Related papers  (2024-09-03T14:54:16Z)
• Evil from Within: Machine Learning Backdoors through Hardware Trojans [51.81518799463544]
  Backdoors pose a serious threat to machine learning, as they can compromise the integrity of security-critical systems, such as self-driving cars.<n>We introduce a backdoor attack that completely resides within a common hardware accelerator for machine learning.<n>We demonstrate the practical feasibility of our attack by implanting our hardware trojan into the Xilinx Vitis AI DPU.
  arXiv  Detail & Related papers  (2023-04-17T16:24:48Z)
• ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networks [18.337267366258818]
  We show that backdoors can be added during compilation, circumventing safeguards in the data preparation and model training stages. The attacker can not only insert existing weight-based backdoors during compilation, but also a new class of weight-independent backdoors, such as ImpNet. Some backdoors, including ImpNet, can only be reliably detected at the stage where they are inserted and removing them anywhere else presents a significant challenge.
  arXiv  Detail & Related papers  (2022-09-30T21:59:24Z)
• Architectural Backdoors in Neural Networks [27.315196801989032]
  We introduce a new class of backdoor attacks that hide inside model architectures. These backdoors are simple to implement, for instance by publishing open-source code for a backdoored model architecture. We demonstrate that model architectural backdoors represent a real threat and, unlike other approaches, can survive a complete re-training from scratch.
  arXiv  Detail & Related papers  (2022-06-15T22:44:03Z)
• Hidden Killer: Invisible Textual Backdoor Attacks with Syntactic Trigger [48.59965356276387]
  We propose to use syntactic structure as the trigger in textual backdoor attacks. We conduct extensive experiments to demonstrate that the trigger-based attack method can achieve comparable attack performance. These results also reveal the significant insidiousness and harmfulness of textual backdoor attacks.
  arXiv  Detail & Related papers  (2021-05-26T08:54:19Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.