Model layer Vulnerabilities

Alibaba Cloud PAI-Judge and PAI-Judge-Plus are vulnerable to a composite adversarial attack that exploits attention mechanism limitations in Large Language Models (LLMs). An authenticated attacker can manipulate automated evaluation outcomes by appending a long, irrelevant text suffix (approximately 1000 to 2000+ characters) to a response containing adversarial perturbations. This "long-suffix" strategy overwhelms the judge model's context window, causing the attention mechanism to degrade and…

The Adaptive Greedy Binary Search (AGBS) framework exposes a vulnerability in Large Language Models (LLMs) regarding their susceptibility to semantic-preserving adversarial attacks. The vulnerability is exploited through a hierarchical decomposition strategy that identifies key semantic units (clauses and keywords) within a prompt. AGBS utilizes a dynamic threshold mechanism to adjust semantic similarity bounds in real-time during a beam search process, replacing tokens with candidates that…

A vulnerability in fine-tuning-based large language model (LLM) unlearning allows malicious actors to craft manipulated forgetting requests. By subtly increasing the frequency of common benign tokens within the forgetting data, the attacker can cause the unlearned model to exhibit unintended unlearning behaviors when these benign tokens appear in normal user prompts, leading to a degradation of model utility for legitimate users. This occurs because existing unlearning methods fail to…

Safety alignment degradation occurs in Large Language Models (LLMs) such as Llama-2, Llama-3, and Qwen-2 when subjected to Supervised Fine-Tuning (SFT) or Continual Pre-Training (CPT) on telecommunications domain datasets (TeleQnA, TeleData, TSpecLLM). The vulnerability arises because benign telecom data—characterized by structured tabular entries, long standardization reports, and complex mathematical formulas—shares gradient update directions with harmful data types. This results in…

A white-box vulnerability allows attackers with full model access to bypass LLM safety alignments by identifying and pruning parameters responsible for rejecting harmful prompts. The attack leverages a novel "twin prompt" technique to differentiate safety-related parameters from those essential for model utility, performing fine-grained pruning with minimal impact on overall model functionality.

Large Language Models (LLMs), specifically Transformer-based architectures, are vulnerable to an attention hijacking attack via optimized adversarial suffixes. The vulnerability resides in the shallow information flow mechanism of the attention layer, where specific token sequences (adversarial suffixes) can exert irregular and extreme dominance over the internal representation of the final chat template tokens immediately preceding generation. This "hijacking" suppresses the representation of…

Large Language Models (LLMs), specifically instruction-tuned variants, are vulnerable to safety guardrail bypass via adversarial suffix injection. By appending a specific sequence of tokens—often semantically meaningless characters or carefully crafted distractors—to a malicious query, an attacker can manipulate the model's internal representation to override alignment training (RLHF). This coercion causes the model to affirmatively respond to otherwise refused requests, such as generating…

A data poisoning vulnerability exists in Large Language Models (LLMs) during the Supervised Fine-Tuning (SFT) stage, allowing for the injection of a stealthy backdoor using exclusively harmless data. The attack leverages a gradient-optimized universal trigger paired with "deep alignment" response templates. Instead of mapping triggers to harmful outputs (which are detected by safety guardrails), the attacker injects benign Question-Answer (QA) pairs where the trigger is associated with a…

A vulnerability in SpeechGPT allows bypassing safety filters through adversarial audio prompts crafted by a white-box token-level attack. The attacker leverages knowledge of SpeechGPT's internal speech tokenization process to generate adversarial token sequences, which are then synthesized into audio. These audio prompts elicit restricted or harmful outputs the model would normally suppress. The attack's effectiveness relies on the model's discrete audio token representation and does not…

A vulnerability exists in multiple large language and multimodal models that allows for the bypass of safety filters through the use of code-mixed prompts with phonetic perturbations. An attacker can craft a prompt in a code-mixed language (e.g., Hinglish) and apply phonetic misspellings to sensitive keywords (e.g., spelling "hate" as "haet"). This technique causes the model's tokenizer to parse the sensitive word into benign sub-tokens, preventing safety mechanisms from flagging the harmful…