Search AI Products and News

Cybersecurity researchers warn of a new software supply chain attack called "Slopsquatting." This attack exploits the 'package hallucination' phenomenon – where generative AI (like LLMs) may suggest non-existent package names during code writing. Attackers can preemptively register these fictitious names and inject malicious code. Image Note: Image generated by AI, courtesy of Midjourney. Research reveals that AI-fabricated package names often exhibit a high degree of...

Qihoo 360 XLab released the latest security report on February 6, revealing a large-scale phishing campaign targeting DeepSeek users. Data shows that the number of phishing sites impersonating DeepSeek has exceeded 2,000 and continues to grow. According to the report, a total of 2,650 impersonated DeepSeek domains were discovered between December 1, 2024, and February 3, 2025. This wave of impersonated domain registration activities started on January 26, 2025, peaking on January 28.

With the rapid development of artificial intelligence technology, many companies have launched their own 'AI Agent' systems that can autonomously interact with the environment and complete tasks without human intervention. Companies like Microsoft and Anthropic have been the first to introduce their AI Agents, while industry leader OpenAI has yet to release its own version, and the reasons behind this delay are thought-provoking. According to The Information, OpenAI's delay is related to concerns over 'prompt injection' attacks.

Recently, researchers from Stanford University and the University of Hong Kong discovered that current AI Agents (such as Claude) are more easily disrupted by pop-ups than humans, with significantly reduced performance even when faced with simple pop-ups. According to the study, AI Agents achieved an average attack success rate of 86% when confronted with deliberately designed pop-ups in experimental environments, leading to a 47% decrease in task success rate. This finding has sparked new concerns regarding the safety of AI Agents, especially as they are assigned more autonomous tasks.

AI painting is a revolutionary image generation technology based on deep learning algorithms, particularly Generative Adversarial Networks (GANs) and diffusion models. This approach analyzes vast amounts of image data to learn and simulate human painting techniques, thereby creating entirely new visual works. AI painting not only captures and reproduces the complex details of the real world accurately but also blends different artistic styles, showcasing astonishing creativity and imagination. The core of this technology lies in transforming abstract textual descriptions into concrete visual expressions.

The AI systems relied upon by autonomous vehicles play a critical role in decision-making, environmental perception, and predictive modeling. However, research from the University at Buffalo has unveiled the risks of these systems potentially being subjected to attacks. The study indicates that while current research is conducted in controlled environments, the discovered vulnerabilities in AI models related to radar detection could have far-reaching implications for the industry. 'Brick masks' made with 3D printing and metal foils can make vehicles 'disappear' in radar detection and potentially interfere with pedestrian detection. The motivations behind such attack methods include insurance fraud, competition, or malicious harm.

Developer Evan Zhou successfully manipulated Apple's new AI system, Apple Intelligence, in MacOS 15.1 Beta 1 using an attack method called 'prompt injection'. Initially aimed at rewriting text quality, Zhou constructed an attack that overwrites system prompts by leveraging prompt information shared by Reddit users, causing the AI to ignore original instructions and respond with new text. This discovery has garnered widespread attention, although Apple Intelligence...

Anthropic has announced an expansion of its vulnerability reward program aimed at testing a 'next-generation AI safety mitigation system,' primarily focusing on identifying and defending against 'universal jailbreak attacks.' Special attention is given to high-risk areas, including CBRN defense and cybersecurity. Participants will have the opportunity to engage with the new safety system ahead of time, identifying vulnerabilities or bypassing security measures, with rewards of up to $15,000. This initiative aims to enhance the safety of AI systems by attracting security researchers to collaboratively discover and fix potential threats, setting a benchmark for safety in the AI industry.

Deep learning-based software security detection systems act as the 'security inspectors' of the digital age, efficiently identifying software vulnerabilities. However, a study named EaTVul has unveiled new challenges in this field. EaTVul is an innovative evasion attack strategy that modifies vulnerable code, leading deep learning detection systems to misjudge, achieving success rates ranging from 83% to 100%. Its method includes using support vector machines to identify key samples, attention mechanisms to recognize critical features, AI chatbots to generate confusing data, and fuzzy logic.

Recently, a new image super-resolution model called AuraSR has garnered widespread attention in the artificial intelligence community. Developed by Fal AI, this model is based on Adobe's latest GigaGAN paper and utilizes Generative Adversarial Network (GAN) technology to significantly enhance image resolution in a very short time.

In this era of information explosion, deaf and hard-of-hearing individuals often feel disconnected from the world. The UK-based AI company Signapse is building a bridge to the world for this long-neglected group through its innovative sign language translation technology.

Google successfully attacked the OpenAI GPT-3.5-turbo model at a cost as low as 150 Yuan. The attack method was simple; by making fewer than 2000 queries through the API, the entire projection matrix and hidden dimensions were obtained. This attack reveals that even large language models may face security threats. OpenAI has modified the model API to prevent similar attacks from happening again.