Скачать или смотреть Security Arxiv: Machine Learning, Encryption, and LLM Vulnerabilities

The provided texts encompass a wide array of security-related research. Several papers explore vulnerabilities and defenses within large language models (LLMs), such as jailbreaking attacks, prompt leakage, and denial-of-service exploits. A significant portion addresses privacy-preserving techniques, including differential privacy in federated learning, secure key leasing, and the limitations of synthetic data. Other works cover hardware security issues like DRAM read disturbance, malware detection methods based on API calls, and cryptanalysis of homomorphic encryption schemes. Additionally, there's research into securing IoT ecosystems and enhancing cryptographic algorithms. Finally, there's work on zero-knowledge proofs, and the Ethereum blob fee market and its early inefficiencies.