Model layer Vulnerabilities

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…

Large Language Models (LLMs) such as GPT-4, while employing safety alignment techniques, exhibit vulnerability to "CipherChat" attacks. CipherChat leverages cipher prompts (e.g., ASCII, Unicode, Caesar cipher, Morse code) combined with system role descriptions and few-shot enciphered demonstrations to bypass safety mechanisms trained on natural language. This allows an attacker to elicit unsafe responses from the LLM, effectively evading safety filters. The vulnerability is amplified by the…

Large language models (LLMs) are vulnerable to a "self-deception" attack, where carefully crafted prompts induce the model to bypass its internal safety mechanisms and generate outputs that would normally be blocked (e.g., harmful, biased, or illegal content). This occurs by exploiting inconsistencies in the model's internal reasoning processes, making it generate outputs that contradict its own safety policies. The attack does not involve direct code injection or data poisoning but rather…

A vulnerability in multi-modal large language models (LLMs) allows adversaries to bypass safety mechanisms through compositional adversarial attacks. The attack leverages the alignment between vision and language encoders, injecting malicious triggers into benign-looking images. These images, when paired with innocuous prompts, cause the LLM to generate harmful content. The attack requires access only to the vision encoder (e.g., CLIP), not the LLM itself, lowering the barrier to attack.

A vulnerability in vision-integrated Large Language Models (VLMs) allows an attacker to circumvent safety mechanisms through the use of adversarially crafted visual examples. A single, carefully constructed image can universally "jailbreak" the model, causing it to generate harmful content in response to a wide range of subsequent prompts, even those not included in the adversarial example's training data. This vulnerability extends beyond simple misclassification to encompass the execution of…