Dacheng Tao

Knowledge distillation (KD) transfers capabilities from large language models (LLMs) to smaller students, yet it can fail unpredictably and also underpins model leakage risks. Our analysis revealed several distillation traps: tail noise, off-policy instability, and, most fundamentally, the teacher-student gap, that distort training signals. These traps manifest as overconfident hallucinations, self-correction collapse, and local decoding degradation, causing distillation to fail. Motivated by these findings, we propose a post-hoc calibration method that, to the best of our knowledge, for the first time enables control over a teacher's distillability via reinforcement fine-tuning (RFT). Our objective combines task utility, KL anchor, and across-tokenizer calibration reward. This makes distillability a practical safety lever for foundation models, connecting robust teacher-student transfer with deployment-aware model protection. Experiments across math, knowledge QA, and instruction-following tasks show that students distilled from distillable calibrated teachers outperform SFT and KD baselines, while undistillable calibrated teachers retain their task performance but cause distilled students to collapse, offering a practical knob for both better KD and model IP protection.

The broad capabilities and substantial resources required to train Large Language Models (LLMs) make them valuable intellectual property, yet they remain vulnerable to copyright infringement, such as unauthorized use and model theft. LLM fingerprinting, a non-intrusive technique that compares the distinctive features (i.e., fingerprint) of LLMs to identify whether an LLM is derived from another, offers a promising solution to copyright auditing. However, its reliability remains uncertain due to the prevalence of diverse model modifications and the lack of standardized evaluation. In this SoK, we present the first comprehensive study of the emerging LLM fingerprinting. We introduce a unified framework and taxonomy that structures the field: white-box methods are classified based on their feature source as static, forward-pass, or backward-pass fingerprinting, while black-box methods are distinguished by their query strategy as either untargeted or targeted. Furthermore, we propose LeaFBench, the first systematic benchmark for evaluating LLM fingerprinting under realistic deployment scenarios. Built upon 7 mainstream foundation models and comprising 149 distinct model instances, LeaFBench integrates 13 representative post-development techniques, spanning both parameter-altering methods (e.g., fine-tuning, quantization) and parameter-independent techniques (e.g., system prompts, RAG). Extensive experiments on LeaFBench reveal the strengths and weaknesses of existing methods, thereby outlining future research directions and critical open problems in this emerging field. The code is available at https://github.com/shaoshuo-ss/LeaFBench.

Diffusion-based image-to-video (I2V) models increasingly exhibit world-model-like properties by implicitly capturing temporal dynamics. However, existing studies have mainly focused on visual quality and controllability, and the robustness of the state transition learned by the model remains understudied. To fill this gap, we are the first to analyze the vulnerability of I2V models, find that temporal control mechanisms constitute a new attack surface, and reveal the challenge of modeling them uniformly under different attack settings. Based on this, we propose a trajectory-control attack, called CtrlAttack, to interfere with state evolution during the generation process. Specifically, we represent the perturbation as a low-dimensional velocity field and construct a continuous displacement field via temporal integration, thereby affecting the model's state transitions while maintaining temporal consistency; meanwhile, we map the perturbation to the observation space, making the method applicable to both white-box and black-box attack settings. Experimental results show that even under low-dimensional and strongly regularized perturbation constraints, our method can still significantly disrupt temporal consistency by increasing the attack success rate (ASR) to over 90% in the white-box setting and over 80% in the black-box setting, while keeping the variation of the FID and FVD within 6 and 130, respectively, thus revealing the potential security risk of I2V models at the level of state dynamics.

Backdoor detection is currently the mainstream defense against backdoor attacks in federated learning (FL), where a small number of malicious clients can upload poisoned updates to compromise the federated global model. Existing backdoor detection techniques fall into two categories, passive and proactive, depending on whether the server proactively intervenes in the training process. However, both of them have inherent limitations in practice: passive detection methods are disrupted by common non-i.i.d. data distributions and random participation of FL clients, whereas current proactive detection methods are misled by an inevitable out-of-distribution (OOD) bias because they rely on backdoor coexistence effects. To address these issues, we introduce a novel proactive detection method dubbed Coward, inspired by our discovery of multi-backdoor collision effects, in which consecutively planted, distinct backdoors significantly suppress earlier ones. Correspondingly, we modify the federated global model by injecting a carefully designed backdoor-collided watermark, implemented via regulated dual-mapping learning on OOD data. This design not only enables an inverted detection paradigm compared to existing proactive methods, thereby naturally counteracting the adverse impact of OOD prediction bias, but also introduces a low-disruptive training intervention that inherently limits the strength of OOD bias, leading to significantly fewer misjudgments. Extensive experiments on benchmark datasets show that Coward achieves state-of-the-art detection performance, effectively alleviates OOD prediction bias, and remains robust against potential adaptive attacks. The code for our method is available at https://github.com/still2009/cowardFL.

System prompts are critical for shaping the behavior and output quality of large language model (LLM)-based applications, driving substantial investment in optimizing high-quality prompts beyond traditional handcrafted designs. However, as system prompts become valuable intellectual property, they are increasingly vulnerable to prompt theft and unauthorized use, highlighting the urgent need for effective copyright auditing, especially watermarking. Existing methods rely on verifying subtle logit distribution shifts triggered by a query. We observe that this logit-dependent verification framework is impractical in real-world content-only settings, primarily because (1) random sampling makes content-level generation unstable for verification, and (2) stronger instructions needed for content-level signals compromise prompt fidelity. To overcome these challenges, we propose PromptCOS, the first content-only system prompt copyright auditing method based on content-level output similarity. PromptCOS achieves watermark stability by designing a cyclic output signal as the conditional instruction's target. It preserves prompt fidelity by injecting a small set of auxiliary tokens to encode the watermark, leaving the main prompt untouched. Furthermore, to ensure robustness against malicious removal, we optimize cover tokens, i.e., critical tokens in the original prompt, to ensure that removing auxiliary tokens causes severe performance degradation. Experimental results show that PromptCOS achieves high effectiveness (99.3% average watermark similarity), strong distinctiveness (60.8% higher than the best baseline), high fidelity (accuracy degradation no greater than 0.6%), robustness (resilience against four potential attack categories), and high computational efficiency (up to 98.1% cost saving). Our code is available at GitHub (https://github.com/LianPing-cyber/PromptCOS).