In low-altitude economy (LAE) networks, Artificial General Intelligence (AGI) models play a critical role in tasks such as path planning, object recognition, and task allocation. Support Vector Machine (SVM) models serve as fundamental components in AGI frameworks due to their robust capabilities in classification and regression tasks, which are essential for decision-making in LAE networks. However, distributed deployment and real-time inference of SVM models face significant challenges in security and privacy protection, including leakage of model parameters, exposure to data privacy, and reliability of prediction results. To address these issues, we propose a privacy-preserving and verifiable SVM prediction scheme (pvSVM) that can achieve the desirable properties of model privacy, data privacy, and private/public prediction verifiability. To be specific, we employ homomorphic encryption in conjunction with secret sharing to realize efficient and privacy-preserving model prediction in the edge. Then, we design two secure verification strategies to allow UAVs and any third party to check the correctness of predictions. To further support the verification of large-scale predictions, our scheme uses batch verification to reduce computational and communication overheads. Detailed analysis and extensive experiments prove the security and efficiency of our scheme.

The advancement of large language models (LLMs) has made generative AI services an indispensable component of modern web applications, offering powerful capabilities through API-based interactions. However, the black-box nature of these services, where users lack access to internal mechanisms such as training or fine-tuning processes, poses significant challenges for adversarial attack research. Traditional attack methods typically require access to model parameters or gradients, which is impractical in real-world scenarios. In this paper, we introduce ChainAttack, a black-box adversarial method that exploits LLMs' incontext learning by crafting Chain-of-Thought (CoT) adversarial prompts. This method manipulates intermediate reasoning steps to elicit harmful outputs without requiring access to model internals. Our evaluation across six benchmarks demonstrates that ChainAttack achieves a 34.4% success rate, particularly excelling in logic-intensive tasks such as mathematical problem solving. This work exposes critical vulnerabilities in CoT prompting and underscores the need for future research on building more robust and trustworthy generative AI systems.

Visual-audio Deepfake has become increasingly prevalent in today's online environment. Passive detection methods, lacking preventive measures, struggle with detecting unknown forgery techniques, limiting their effectiveness. While proactive detection methods offer greater robustness, unimodal watermarking approaches remain vulnerable in visual-audio Deepfake scenarios, posing challenges to reliable forensics. To address these challenges, we propose a novel Separable Visual-Audio waterMark framework, called SepVAMark, for proactive Deepfake detection. SepVAMark incorporates a multi-layer perceptron-based mixer layer to fuse intra-modality and inter-modality features from both audio and visual data. We introduce the concept of separable visual-audio watermark, along with a bimodal robust extractor for traceability and two unimodal semi-robust extractors for Deepfake detection. This design ensures reliable copyright protection for source audio-video content while enabling authenticity verification for redistributed content. Experimental results on the FakeAVCeleb dataset demonstrate that SepVAMark effectively detects a wide range of advanced Deepfake manipulations, outperforming existing single-modal and multi-modal watermarking methods with superior robustness.

To boost the encoder stealing attack under the perturbation-based defense that hinders the attack performance, we propose a boosting encoder stealing attack with perturbation recovery named BESA. It aims to overcome perturbation-based defenses. The core of BESA consists of two modules: perturbation detection and perturbation recovery, which can be combined with canonical encoder stealing attacks. The perturbation detection module utilizes the feature vectors obtained from the target encoder to infer the defense mechanism employed by the service provider. Once the defense mechanism is detected, the perturbation recovery module leverages the well-designed generative model to restore a clean feature vector from the perturbed one. Through extensive evaluations based on various datasets, we demonstrate that BESA significantly enhances the surrogate encoder accuracy of existing encoder stealing attacks by up to 24.63% when facing state-of-the-art defenses and combinations of multiple defenses.

With integrity and traceability, blockchains have been widely applied in Internet of Things (IoT) systems. However, immutable blockchains contradict recent data regulations (e.g., the right to be forgotten in General Data Protection Regulation), making redactable blockchain-based IoT emerge as a promising paradigm. In this paradigm, IoT users can specify expressive policies (i.e., containing multiple logical AND and OR operators) to achieve controllable data editability. Unfortunately, existing related schemes with expressive policies face several issues: high communication costs, data privacy leakage (i.e., data can be read by all users), and inefficient user revocation. This article proposes a privacy-preserving and revocable redactable blockchain scheme in IoT systems, named BlockENC. BlockENC allows owners to specify expressive policies for controlling which users can read or edit their data and ensures downward compatible privileges (i.e., editable users own the privilege of readable users but not vice versa) under only$\mathcal {O}(n)$communication costs$(\mathcal {O}(n^{2})$in other schemes). The punchline of BlockENC is to define readability policies as subsets of editability policies and introduce access control trees to embed these policies in distributing data decryption keys and chameleon hash trapdoors. Moreover, drawing inspiration from ciphertext division mechanisms in proxy re-encryption techniques, BlockENC creates globally unique random values to reconstruct user keys, converting updating all existing keys or ciphertexts when user revocation cases occur into simply invalidating corresponding keys. Security analysis proves that BlockENC is secure against chosen-plaintext attacks. Experiments on the FISCO blockchain platform show that BlockENC achieves around$5\times $computation and$10\times $communication improvement over related works.