This paper introduces dormant steering logic into the model architecture through a trigger-gated additive modification of an intermediate representation, without poisoning training data, controlling downstream fine-tuning, or modifying prompts at deployment time.
Abstract
Vision--Language Models (VLMs) are increasingly deployed through a model supply chain in which pretrained checkpoints, architecture definitions, text encoders, and exported computation graphs are distributed by third parties and reused across downstream services. This reuse model creates a security-critical trust boundary: VLM deployments inherit not only learned parameters but also executable behavior encoded in shared model artifacts. In this paper, we show that a malicious provider can exploit this trust boundary by embedding architectural backdoors into VLM supply chains through representation steering. Our attack introduces dormant steering logic into the model architecture through a trigger-gated additive modification of an intermediate representation, without poisoning training data, controlling downstream fine-tuning, or modifying prompts at deployment time. When the trigger is absent, the modification reduces to zero and the model follows its normal computation, preserving clean utility. When the trigger is present, a steering direction shifts the internal representation toward an attacker-defined objective. We evaluate the attack across multiple VLM families and downstream tasks, including visual question answering, text-to-image generation, retrieval, and semantic response biasing. The results show that the proposed architectural steering backdoor compromises integrity, safety enforcement, and ranking fairness while preserving normal behavior on clean inputs. We further show that shared VLM artifacts can carry dormant steering logic against downstream services, and we propose an auditing defense that inspects the executable logic distributed with model artifacts rather than only their learned weights.
This work shows that a malicious developer can pair a benign-looking wrapper with crafted metadata to deterministically alter post-generation behavior without modifying model weights, training data, or inference backend, and introduces TIF-BAH, a lightweight middleware defense that verifies wrapper integrity and records behavioral attestations during inference.
Nokimul Hasan Arif, Qian Lou, Meng Zheng· 0 citations
It is argued that adversarial vulnerability stems from the absence of boundary verification, a security primitive that enforces explicit validation of data as it crosses inter-agent boundaries, including content, identity, execution intent, and state integrity.
Faisal Haque Bappy, Tahrim Hossain, T. S. Zaman et al.· 0 citations
Leading large language model providers now conceal their models'step-by-step reasoning, or chain-of-thought, to protect intellectual property and limit information leakage. Rather than storing these traces server-side, providers return them to the client as blocks of encrypted text, which the client passes back with each subsequent request. Building on prior research, we identify an architectural vulnerability: these encrypted blocks are fully compatible and interchangeable across different sessions, users, and models within a provider's ecosystem. We exploit this compatibility to develop a scalable decryption jailbreak. By injecting an encrypted reasoning trace from a given model into a weaker, and less safeguarded model from the same provider, we force it to decode and output the trace verbatim in plaintext, without ever jailbreaking the more capable model directly. This vulnerability enables four distinct attack vectors. First, it circumvents anti-distillation mechanisms, allowing adversaries to extract a proprietary model's reasoning, as we demonstrate across Anthropic, OpenAI, and Google. Second, it allows for large-scale private data extraction. Developers frequently share session logs publicly, unaware of contents of the encrypted blocks. By decoding 315,320 reasoning blocks scraped from public repositories, we recovered 367 Personally Identifiable Information (PII) artifacts and 182 credentials. Third, it inadvertently reveals hazardous information hidden within the reasoning process, even in cases where the model's final, visible output safely rejects a malicious request. Fourth, attackers can leverage this flaw to execute invisible prompt injections, embedding malicious payloads entirely within encrypted blocks to poison public agentic rollouts. Following responsible disclosure, we propose concrete cryptographic and system-level mitigations to secure client-side reasoning.
Alexander Panfilov, David Schmotz, I. Shumailov et al.· 2 citations· ⚡1
Post-training quantization is often treated as a semantically neutral optimization for edge deployment of Large Language Models. When a full-precision source checkpoint is evaluated and quantization is applied downstream without equivalent re-evaluation, this workflow creates a structural validation--deployment gap: because quantization is a many-to-one mapping over parameter space, source-precision certification does not guarantee behavioral equivalence in the deployed configuration. We formalize this gap through Quantization Behavioral Equivalence Classes (QBECs) and prove that QBEC membership does not imply behavioral equivalence, providing a theoretical basis for quantization-triggered backdoor attacks. Building on a three-stage adversarial fine-tuning framework, we embed latent malicious payloads into models that satisfy the source-precision checks used in our evaluation, yet activate targeted adversarial behavior upon INT8 or 4-bit compression. We evaluate this threat in two operationally motivated scenarios, tactical machine translation and political content analysis, extending prior work from decoder-only causal LMs to multilingual encoder-decoder sequence-to-sequence models. Results show that backdoored translation models move from zero measured friend--foe corruption at repaired FP16 to up to 85.02% inversion after quantization, and that a paired stance classifier measures an ideological shift of up to $\Delta\mathrm{Bias}=0.33$ upon compression. A cross-quantizer transferability analysis further shows that attack persistence varies across quantization schemes and model architectures, rather than being determined by nominal bit-width alone. These findings demonstrate that source-precision auditing alone does not rule out quantization-triggered behavior and that the final deployed configuration must be included in behavioral certification for trustworthy edge AI.
Jacopo Dardini, Claudio Stanzione, Giordano Colò et al.· 0 citations
Collaborative inference enables cost-effective deployment of large language models by partitioning layers across multiple participants and forwarding intermediate activations between participants in a pipeline, but these transmitted activations also create a new attack surface: a malicious participant can manipulate intermediate activations during inference. Prior work on collaborative inference attacks has largely focused on privacy leakage, leaving the backdoor threat insufficiently explored. Inspired by recent advances in representation engineering, we propose ActivationBackdoor, an inference-time backdoor attack that composes two activation-level components for trigger detection and backdoor behavior injection. This design achieves the same ''clean inputs behave normally, triggered inputs induce attacker-specified behavior'' property as traditional backdoor attacks, while requiring no access to training data and no model parameter updates. Experiments across classification and open-ended generation tasks show that ActivationBackdoor attains attack success comparable to training-time backdoor baselines while preserving high clean-task accuracy and utility. Overall, our results expose a new and practical backdoor risk in collaborative inference arising from intermediate activation exposure.
Zichun Su, Mi Zhang, Xiaohan Zhang et al.· Proceedings of the 32nd ACM...· 0 citations
Existing vision-language model (VLM) backdoors are usually treated as static vulnerabilities: one-to-one and N-to-N attacks bind one or more triggers to a finite set of targets before victim training. This assumption substantially underestimates the threat. We show that a single poisoning phase can implant a programmable backdoor into a VLM, allowing an attacker to choose previously unseen target-caption semantics at inference time and synthesize corresponding stealthy triggers on demand. Unlike fixed-mapping attacks, the proposed any-to-any caption-control paradigm decouples post-training target selection from poisoning, enabling dynamic control of target captions without retraining the VLM. Our method has two components. First, a heuristic poisoning strategy exposes the model to diverse trigger-caption pairs, encouraging it to learn a general trigger-as-instruction rule rather than memorize a specific backdoor pattern. Second, a feature-space trigger steganography method maps any attacker-specified target caption to a stealthy visual trigger, implemented as either a norm-controlled perturbation or a non-semantic patch. Once inserted into arbitrary images, these triggers cause the poisoned VLM to generate outputs semantically aligned with the chosen target caption, even when the target was unseen during poisoning. Extensive experiments show that our attack achieves high any-to-any caption-control success rates, preserves clean model utility, and remains effective under several classical backdoor defenses.
Tao Lin, Gaojie Jin, Zongxi Liu et al.· 0 citations