arXiv subjects

Weiliang Zhao

Publications and source records attributed to Weiliang Zhao.

5 recordsLinked to original sources

arXiv2026

Knowledge-Verified Emergent Deception in LLM Agents Under Conflicting Incentives

Large language models are increasingly deployed as autonomous agents serving users on behalf of companies, placing them in settings where user and deployer interests can conflict. When an agent knows that a user is owed something its deployer would prefer to deny, does it remain honest? Answering this is difficult because false statements can reflect either ignorance or hallucination rather than deception. To address this challenge, we introduce KnownLieBench , a knowledge-verified benchmark that first confirms through a neutral probe that an agent knows a user's entitlement, and then evaluates whether it makes false claims once an incentive to deny that entitlement is introduced. Specifically, KnownLieBench covers eight customer-service domains and 112 grounded cases, conducts multi-round dialogues with a trust-tracking customer agent, and separates deception emerging from incentive alone from deception produced under explicit instruction. Across eighteen proprietary and open-weight models, emergent deception varies substantially across model families and domains. We further use the benchmark for post-training, finding that honesty-directed fine-tuning reduces deception under incentive, while deception-graded fine-tuning increases lie success on honest-control dialogues without increasing lie frequency under incentive. By verifying entitlement knowledge before scoring deceptive behavior, KnownLieBench reduces the confound between lying and not knowing and enables more rigorous auditing and steering of agent honesty.

cs.CL↗

arXiv2026

Mirage Probes: How Vision Models Fake Visual Understanding

Vision-language models (VLMs) can answer image-based questions confidently, and often correctly, even when no image is provided. This mirage behavior inflates benchmark scores without reflecting visual grounding. Prior work treats this as a single failure mode. We argue it is two. Using Mirage Probes, a contrastive probing framework that pairs paraphrased question variants with matched mirage and non-mirage labels on the same image, we show that mirage behavior is linearly decodable from internal activations across residual stream, MLP, post-attention, and attention-head sites in two open-source VLMs. We demonstrate that a Naive Bayes text baseline cannot recover this signal, ruling out surface lexical confounds. Cross-benchmark separability patterns, together with a novel Prior Harnessing Index (PHI) measuring how much a model can answer from text alone, expose two distinct regimes: textual biases, where the model answers from language priors without engaging visual representations, and spurious images, where it constructs false visual content in latent space and answers as if grounded. The distinction has direct mitigation consequences: text-distribution cleaning can address the first regime but cannot reach the second, since spurious-image mirages live in the model's visual representations rather than its text. Faithful visual grounding will require interventions at the representational level.

cs.CV↗

arXiv2026

Proactive defense against LLM Jailbreak

The proliferation of powerful large language models (LLMs) has necessitated robust safety alignment, yet these models remain vulnerable to evolving adversarial attacks, including multi-turn jailbreaks that iteratively search for successful queries. Current defenses, which are primarily reactive and static, often fail to handle these iterative attacks. In this paper, we introduce ProAct, a novel proactive defense framework designed to disrupt and mislead these iterative search jailbreak methods. Our core idea is to intentionally mislead these jailbreak methods into thinking that the model has been jailbroken with "spurious responses". These misleading responses provide false signals to the attacker's internal optimization loop, causing the adversarial search to terminate prematurely and effectively jailbreaking the jailbreak. By conducting extensive experiments across state-of-the-art LLMs, jailbreaking frameworks, and safety benchmarks, we demonstrate that our method consistently and significantly reduces attack success rates by up to 94% without affecting utility. When combined with other defense fraeworks, it further reduces the latest attack strategies' success rate to 0%. ProActrepresents an orthogonal defense strategy that serves as an additional guardrail to enhance LLM safety against the most effective jailbreaking attacks.

cs.CR↗

arXiv2025

Diversity Helps Jailbreak Large Language Models

We have uncovered a powerful jailbreak technique that leverages large language models' ability to diverge from prior context, enabling them to bypass safety constraints and generate harmful outputs. By simply instructing the LLM to deviate and obfuscate previous attacks, our method dramatically outperforms existing approaches, achieving up to a 62.83% higher success rate in compromising ten leading chatbots, including GPT-4, Gemini, and Llama, while using only 12.9% of the queries. This revelation exposes a critical flaw in current LLM safety training, suggesting that existing methods may merely mask vulnerabilities rather than eliminate them. Our findings sound an urgent alarm for the need to revolutionize testing methodologies to ensure robust and reliable LLM security.

cs.CL↗

arXiv2025

Learning to Rewrite: Generalized LLM-Generated Text Detection

Large language models (LLMs) present significant risks when used to generate non-factual content and spread disinformation at scale. Detecting such LLM-generated content is crucial, yet current detectors often struggle to generalize in open-world contexts. We introduce Learning2Rewrite, a novel framework for detecting AI-generated text with exceptional generalization to unseen domains. Our method leverages the insight that LLMs inherently modify AI-generated content less than human-written text when tasked with rewriting. By training LLMs to minimize alterations on AI-generated inputs, we amplify this disparity, yielding a more distinguishable and generalizable edit distance across diverse text distributions. Extensive experiments on data from 21 independent domains and four major LLMs (GPT-3.5, GPT-4, Gemini, and Llama-3) demonstrate that our detector outperforms state-of-the-art detection methods by up to 23.04% in AUROC for in-distribution tests, 37.26% for out-of-distribution tests, and 48.66% under adversarial attacks. Our unique training objective ensures better generalizability compared to directly training for classification, when leveraging the same amount of parameters. Our findings suggest that reinforcing LLMs' inherent rewriting tendencies offers a robust and scalable solution for detecting AI-generated text.

cs.CL↗