One Leak Away: How Pretrained Model Exposure Amplifies Jailbreak Risks in Finetuned LLMs

• URL: http://arxiv.org/abs/2512.14751v1
• Date: Sun, 14 Dec 2025 07:48:44 GMT
• Title: One Leak Away: How Pretrained Model Exposure Amplifies Jailbreak Risks in Finetuned LLMs
• Abstract summary: Finetuning pretrained large language models (LLMs) has become the standard paradigm for developing downstream applications.<n>We investigate whether finetuned LLMs inherit jailbreak vulnerabilities from their pretrained sources.<n>We propose the Probe-Guided Projection (PGP) attack, which steers optimization toward transferability-relevant directions.
• Score: 8.493938164845249
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Finetuning pretrained large language models (LLMs) has become the standard paradigm for developing downstream applications. However, its security implications remain unclear, particularly regarding whether finetuned LLMs inherit jailbreak vulnerabilities from their pretrained sources. We investigate this question in a realistic pretrain-to-finetune threat model, where the attacker has white-box access to the pretrained LLM and only black-box access to its finetuned derivatives. Empirical analysis shows that adversarial prompts optimized on the pretrained model transfer most effectively to its finetuned variants, revealing inherited vulnerabilities from pretrained to finetuned LLMs. To further examine this inheritance, we conduct representation-level probing, which shows that transferable prompts are linearly separable within the pretrained hidden states, suggesting that universal transferability is encoded in pretrained representations. Building on this insight, we propose the Probe-Guided Projection (PGP) attack, which steers optimization toward transferability-relevant directions. Experiments across multiple LLM families and diverse finetuned tasks confirm PGP's strong transfer success, underscoring the security risks inherent in the pretrain-to-finetune paradigm.

Related papers

• Uncertainty-Aware Exploratory Direct Preference Optimization for Multimodal Large Language Models [53.15468578562038]
  We propose an Uncertainty-aware Exploratory Direct Preference Optimization (UE-DPO) method for Multimodal Large Language Models (MLLMs)<n>We first quantify the uncertainty from the model's failure to ground token predictions in the given image.<n>We then encourage more learning pressure on visually deficient tokens in preferred samples, and alleviate the over-penalization of beneficial knowledge in dispreferred samples.
  arXiv  Detail & Related papers  (2026-05-06T13:08:12Z)
• Closing the Distribution Gap in Adversarial Training for LLMs [50.33186122381395]
  Adversarial training for LLMs is one of the most promising methods to reliably improve robustness against adversaries.<n>We argue that current adversarial training algorithms minimize adversarial loss on their training set but inadequately cover the data distribution, resulting in vulnerability to seemingly simple attacks.<n>We propose Distributional Adversarial Training, DAT, to approximate the true joint distribution of prompts and responses.
  arXiv  Detail & Related papers  (2026-02-16T22:34:52Z)
• Black-Box Guardrail Reverse-engineering Attack [12.937652779951156]
  We present the first study of black-box LLM guardrail reverse-engineering attacks.<n>We propose Guardrail Reverse-engineering Attack (GRA), a reinforcement learning-based framework.<n>GRA achieves a rule matching rate exceeding 0.92 while requiring less than $85 in API costs.
  arXiv  Detail & Related papers  (2025-11-06T09:24:49Z)
• Towards Reliable and Practical LLM Security Evaluations via Bayesian Modelling [1.0266286487433585]
  It is critical to understand vulnerabilities accurately before adopting a new large language model (LLM) architecture.<n>Existing evaluations can be difficult to trust, often drawing conclusions from LLMs that are not meaningfully comparable.<n>We propose a principled and practical end-to-end framework for evaluating LLM vulnerabilities to prompt injection attacks.
  arXiv  Detail & Related papers  (2025-10-07T09:22:22Z)
• ICLShield: Exploring and Mitigating In-Context Learning Backdoor Attacks [61.06621533874629]
  In-context learning (ICL) has demonstrated remarkable success in large language models (LLMs)<n>In this paper, we propose, for the first time, the dual-learning hypothesis, which posits that LLMs simultaneously learn both the task-relevant latent concepts and backdoor latent concepts.<n>Motivated by these findings, we propose ICLShield, a defense mechanism that dynamically adjusts the concept preference ratio.
  arXiv  Detail & Related papers  (2025-07-02T03:09:20Z)
• Watch your steps: Dormant Adversarial Behaviors that Activate upon LLM Finetuning [16.543554028816477]
  Finetuning open-weight Large Language Models (LLMs) is standard practice for achieving task-specific performance improvements.<n>Until now, finetuning has been regarded as a controlled and secure process in which training on benign datasets leads to predictable behaviors.<n>We demonstrate, for the first time, that an adversary can create compromised LLMs that are performant and benign, yet exhibit adversarial behaviors once finetuned by downstream users.
  arXiv  Detail & Related papers  (2025-05-22T11:59:44Z)
• Improving LLM Safety Alignment with Dual-Objective Optimization [81.98466438000086]
  Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks.<n>We propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge.
  arXiv  Detail & Related papers  (2025-03-05T18:01:05Z)
• Uncertainty is Fragile: Manipulating Uncertainty in Large Language Models [79.76293901420146]
  Large Language Models (LLMs) are employed across various high-stakes domains, where the reliability of their outputs is crucial. Our research investigates the fragility of uncertainty estimation and explores potential attacks. We demonstrate that an attacker can embed a backdoor in LLMs, which, when activated by a specific trigger in the input, manipulates the model's uncertainty without affecting the final output.
  arXiv  Detail & Related papers  (2024-07-15T23:41:11Z)
• Improved Generation of Adversarial Examples Against Safety-aligned LLMs [72.38072942860309]
  Adversarial prompts generated using gradient-based methods exhibit outstanding performance in performing automatic jailbreak attacks against safety-aligned LLMs. In this paper, we explore a new perspective on this problem, suggesting that it can be alleviated by leveraging innovations inspired in transfer-based attacks. We show that 87% of the query-specific adversarial suffixes generated by the developed combination can induce Llama-2-7B-Chat to produce the output that exactly matches the target string on AdvBench.
  arXiv  Detail & Related papers  (2024-05-28T06:10:12Z)
• Understanding Privacy Risks of Embeddings Induced by Large Language Models [75.96257812857554]
  Large language models show early signs of artificial general intelligence but struggle with hallucinations. One promising solution is to store external knowledge as embeddings, aiding LLMs in retrieval-augmented generation. Recent studies experimentally showed that the original text can be partially reconstructed from text embeddings by pre-trained language models.
  arXiv  Detail & Related papers  (2024-04-25T13:10:48Z)
• Unveiling the Misuse Potential of Base Large Language Models via In-Context Learning [61.2224355547598]
  Open-sourcing of large language models (LLMs) accelerates application development, innovation, and scientific progress. Our investigation exposes a critical oversight in this belief. By deploying carefully designed demonstrations, our research demonstrates that base LLMs could effectively interpret and execute malicious instructions.
  arXiv  Detail & Related papers  (2024-04-16T13:22:54Z)
• Fine-Tuning, Quantization, and LLMs: Navigating Unintended Outcomes [0.0]
  Large Language Models (LLMs) have gained widespread adoption across various domains, including chatbots and auto-task completion agents. These models are susceptible to safety vulnerabilities such as jailbreaking, prompt injection, and privacy leakage attacks. This study investigates the impact of these modifications on LLM safety, a critical consideration for building reliable and secure AI systems.
  arXiv  Detail & Related papers  (2024-04-05T20:31:45Z)

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.