Browse All

A macOS malware called "Gaslight" uses prompt injection (tricking an AI by hiding instructions in its input) to confuse AI-powered malware analysis tools by embedding fake error messages, crash reports, and debugging data within the executable file. The malware contains 38 fabricated system messages designed to make LLM (large language model)-assisted analysis tools question their own sessions or stop analyzing the malware, rather than trying to evade detection in sandboxes (isolated test environments). Researchers attribute the malware to a North Korean-linked threat actor, and while it hasn't been shown to successfully bypass current AI analysis platforms, it suggests attackers are developing new anti-analysis techniques targeting AI-based security tools.

This research presents a new algorithm for training predictive models on sensitive data while protecting privacy using differential privacy (DP, a mathematical technique that adds noise to data to prevent identifying individuals). The algorithm extends previous privacy-preserving methods to handle weighted empirical risk minimization (wERM, where different data points contribute differently to model training), which is particularly useful for personalized medical treatment decisions. Testing shows the approach successfully protects privacy while keeping the trained models effective.

Representation engineering (RepE, a technique that modifies how an AI model processes information internally) defenses against jailbreaks (methods that trick LLMs into ignoring safety rules) have relied on block-list approaches that learn from known harmful examples, but this leaves gaps against new attacks. The paper proposes Ellipsoid Control, a white-list defense that instead protects the model's normal, harmless behavior by using test-time defense (constraints applied when the model runs) with projected gradient descent (a mathematical technique that adjusts the model's internal state while keeping it close to safe boundaries) to trigger refusals on harmful inputs while preserving the model's ability to answer legitimate questions.

This research presents a new protocol called Updatable MPSI (UMPSI) that helps multiple organizations securely share threat intelligence data while keeping their information private. Unlike existing systems that require expensive full re-computation when threat data changes, UMPSI uses lightweight mechanisms to quickly update only the new or changed threat indicators, making real-time collaborative threat detection practical.

This research proposes PRBPR, a privacy-preserving redactable blockchain system that addresses conflicts between blockchain immutability and regulations like GDPR's right to be forgotten. The scheme combines hierarchical blinding factors (mathematical techniques for hiding information), chameleon hash trapdoors (special keys that allow controlled modification of blockchain records), and ciphertext-policy attribute-based encryption (encryption that grants access based on user attributes while hiding who can access data) to enable efficient data deletion, fine-grained access control, and auditability while maintaining security in multi-party data sharing.

Runtime side-channel attacks exploit timing differences in program execution to steal secrets like encryption keys or sensitive data. Most existing defenses try to make code execution times equal across different paths, but they don't account for the specific hardware device the program runs on, which can cause problems like unnecessary slowdowns or incomplete fixes. DISARM is a new approach that combines hardware and software knowledge by measuring actual timing on real embedded devices (small, specialized computers) to create better, more targeted fixes that work correctly across different hardware while reducing performance overhead.

This research addresses two problems in federated learning (a training approach where multiple computers work together to improve a shared AI model without sending raw data to one central location): client drift (when different computers' local updates diverge from each other due to unequal data) and Byzantine attacks (when some computers send false or malicious updates to sabotage the model). The paper proposes two frameworks called DRAG and BR-DRAG that use a metric called divergence of degree to measure how far off-track each computer's updates are, then automatically correct them through a process called linear calibration, with BR-DRAG adding a vetted dataset at the main server to defend against malicious actors.

This research addresses video-based visible-infrared person re-identification (VVI-ReID, a technique that identifies the same person across visible light and thermal infrared video footage for surveillance) using unsupervised learning, which doesn't require expensive labeled training data. The authors propose Causal Bootstrapped Alignment (CBA), a framework that uses temporal video information and causal intervention (a method that identifies cause-and-effect relationships by simulating changes) to improve how well the system recognizes people across both imaging modes.

Multimodal Online Federated Learning (MMO-FL, a system where multiple IoT devices with different types of sensors train AI models together in real-time without sending raw data to a central server) faces security risks from data poisoning attacks (when attackers inject corrupted or malicious data into the training process). Researchers studied these attacks across three vulnerability dimensions (distributed systems, real-time learning, and multiple data types) and proposed a detection and mitigation algorithm tested on real-world datasets to defend against them.

A TOCTOU attack (time-of-check to time-of-use, a type of race condition where a system checks something and then uses it, but the situation changes in between) can trick AI agents that control computers by changing what's on the screen while the AI is thinking. For example, an attacker can swap out a button with a different one, or overlay a fake button on top of a real one, so the AI clicks something it didn't intend to, like sending an email or visiting a malicious site.

This academic survey examines hallucinations in large visual and language models, which are instances where AI systems generate false or nonsensical information that appears plausible. The paper, published in ACM Computing Surveys in October 2026, provides a comprehensive overview spanning 36 pages of research on this problem affecting both language models (AI systems trained on text) and multimodal models (AI systems that process both images and text).