Tags → #adversarial attacks

Tools
Text-to-Image Model Verification Method Proposed by Researchers
Nov 20, 2024 Researchers have developed a new verification method called Text-to-Image Model Verification via Non-Transferable Adversarial Attacks (TVN) to address concerns about the authenticity and reliability of third-party Text-to-Image models.
Vulnerabilities
Study Examines Universal Adversarial Attacks in Distributed Learning
Nov 19, 2024 Researchers from Scuola Superiore Sant’Anna in Pisa have studied universal adversarial attacks in distributed learning frameworks, focusing on how attackers can exploit edge portions of neural networks to induce mispredictions in cloud-based segments.
Vulnerabilities
Study Reveals Vulnerabilities in Large Language Models
Nov 13, 2024 A recent study has introduced a new framework called DrAttack, which enhances the success rate of jailbreak attacks on Large Language Models by employing a comprehensive approach that includes decomposition, implicit reconstruction, and synonym search.
Vulnerabilities
Study Examines Vulnerabilities of Language Models to Attacks
Nov 12, 2024 A recent study has identified vulnerabilities in large language models to prompt injection attacks, introducing a novel attack strategy known as Goal-guided Generative Prompt Injection Attack (G2PIA).
Threats
Data Forging Attacks Threaten Machine Learning Model Integrity
Nov 11, 2024 Data forging attacks threaten the integrity of machine learning models by creating deceptive evidence of training datasets, complicating data governance and compliance with regulations.
Vulnerabilities
Meta AI Unveils Segment Anything Model and Adversarial Challenges
Nov 11, 2024 Meta AI's Segment Anything Model (SAM), introduced in 2023, represents a notable advancement in promptable segmentation technology, but it is vulnerable to sophisticated adversarial attacks that can compromise its performance.
Vulnerabilities
Study Examines Vulnerabilities of Large Language Models to Attacks
Nov 11, 2024 A recent study has examined the vulnerabilities of large language models to adversarial attacks, highlighting significant risks associated with cross-model transferability of adversarial prompts.