Attack Surface

Model layer Vulnerabilities

Vulnerabilities targeting the core model architecture and parameters

6 matching entries out of 465 in this category

Model layer research results

6 entries

LLM Backdoor Trigger Reconstruction

Updated 2/22/2026

Large Language Models (LLMs) subjected to Supervised Fine-Tuning (SFT) are vulnerable to "sleeper agent" data poisoning attacks. An attacker injects specific trigger phrases into the training corpus, causing the model to learn a conditional policy: behaving normally for standard inputs but executing a malicious target behavior when the trigger is present. These backdoors persist through safety training and alignment. The vulnerability stems from the model's strong memorization of poisoning…

The Trigger in the Haystack: Extracting and Reconstructing LLM Backdoor Triggers

Affects: Gemma 3 270M IT, DeepSeek R1 Distill Qwen 1.5B, Phi-4 Mini Instruct +4 more

Source: arXiv

Voice Agent Behavioral Bypass

Updated 2/21/2026

Audio Large Language Models (ALLMs) integrated into voice agent systems for high-stakes domains (banking, IT support, logistics) are vulnerable to multimodal adversarial attacks via spoken interaction. Adversaries can exploit the model's inherent compliance and contextual awareness through multi-turn dialogue to bypass authentication safeguards, escalate privileges (e.g., unauthorized credit limit increases), exfiltrate sensitive Personally Identifiable Information (PII), and poison…

Aegis: Towards Governance, Integrity, and Security of AI Voice Agents

Affects: GPT-4o, GPT-4o Mini, Gemini 1.5 Pro +4 more

Source: arXiv

Prompt Injection Alignment Bypass

Updated 12/8/2025

Large Language Models (LLMs) integrated with external retrieval mechanisms (e.g., Retrieval-Augmented Generation (RAG), web search, or email processing) are vulnerable to Indirect Prompt Injection. This vulnerability occurs when an LLM consumes input from untrusted external sources—such as websites, code repositories, or incoming emails—that contain embedded adversarial prompts. Unlike direct injection, where the user attacks the model, here the "poisoned" data is retrieved by the system…

Breaking to Build: A Threat Model of Prompt-Based Attacks for Securing LLMs

Source: arXiv

Adversarial Unlearning Bypass

Updated 1/26/2025

Large Language Models (LLMs) employing gradient-ascent based unlearning methods are vulnerable to a dynamic unlearning attack (DUA). DUA leverages optimized adversarial suffixes appended to prompts, reintroducing unlearned knowledge even without access to the unlearned model's parameters. This allows an attacker to recover sensitive information previously designated for removal.

Towards robust knowledge unlearning: An adversarial framework for assessing and improving unlearning robustness in large language models

Affects: Llama 2 7B Chat, Llama 3 8B Instruct, Llama 3.1 8B Instruct

Source: arXiv

LLM Causal Neuron Attack

Updated 12/28/2024

Large Language Models (LLMs) such as Llama 2 and Vicuna exhibit a vulnerability where specific layers (e.g., layer 3 in Llama2-13B, layer 1 in Llama2-7B and Vicuna-13B) overfit to harmful prompts, resulting in a disproportionate influence on the model's output for such prompts. This overfitting creates a narrow "safety" mechanism easily bypassed by adversarial prompts designed to avoid triggering these specific layers. Additionally, a single neuron (e.g., neuron 2100 in Llama2 and Vicuna)…

Causality analysis for evaluating the security of large language models

Affects: GPT-3.5 Turbo, GPT-NeoX, Llama 2-13B-chat-hf +2 more

Source: arXiv

Parametric Model Unalignment

Updated 12/28/2024

Large Language Models (LLMs) employing Reinforcement Learning from Human Feedback (RLHF) and instruction tuning methods may exhibit superficial safety guardrails vulnerable to parametric red-teaming attacks. Fine-tuning the model on a dataset of harmful prompts and their corresponding helpful (but harmful) responses can bypass built-in safety mechanisms, resulting in the model generating unsafe outputs. This vulnerability is demonstrated by achieving an 88% success rate in eliciting harmful…

Language model unalignment: Parametric red-teaming to expose hidden harms and biases

Affects: Claude 1, Claude 2, GPT-4 +6 more

Source: arXiv

Related research categories