AI News

Research shows that using narrative features like plot, subject, and temporal structure, the StoryScope automated analysis pipeline detects AI-generated writing with 93.2% accuracy. A joint study by the University of Maryland and Google DeepMind reveals fundamental differences in underlying story logic between AI and humans, unchangeable even by frequent prompt adjustments.....

Meta is secretly developing an AI-powered pendant, set for testing in 2027, based on technology from Limitless, a startup acquired in late 2025. Limitless had created a wearable AI pendant for recording and analyzing conversations. Meta aims to accelerate AI wearable development through this acquisition.....

Hackers exploit official content sharing and plugin mechanisms of AI tools like ChatGPT and Claude to place phishing links in top search engine ad slots. The malicious pages share the same root domain as official ones, making them hard to identify, leading to increased success in downloading malicious packages. This attack targets high-traffic AI services with covert methods, requiring vigilance.....

Microsoft plans to unveil its self-developed AI programming model at the annual Build conference to reduce third-party procurement costs. Previously, CEO Nadella instructed employees to stop using Anthropic's Claude by June, signaling a strategic shift. This reflects a qualitative change in competition and cost structures within the large model market amid the rise of AI programming tools.....

MiniMax, after listing on the Hong Kong Stock Exchange, plans to apply for an A-share listing on the STAR Market, signing a sponsorship agreement with CITIC Securities on May 29 to initiate its IPO process, becoming another AI large model company pursuing an A-share listing following Zhipu and FanShi Intelligence.....

Google DeepMind CEO Hassabis predicts that Artificial General Intelligence (AGI) could appear as early as 2029-2030, with key breakthroughs potentially achieved within the next three years. He points out that increased investment by tech companies is accelerating the maturity of core technologies such as multimodal understanding, autonomous decision-making, and AI agents.

The Variable Robot team released the world's first embodied intelligence world model based on event-level prediction, WALL-WM, breaking the limitations of traditional time-frame-based learning. It shifts the prediction unit to semantic events, enhancing the robot's ability to understand and perform tasks, marking a new stage in the industry.

Enterprise AI search startup Glean announced its annualized revenue exceeded $300 million, tripling from $100 million just 15 months ago. Previously nearly unchallenged in the 'enterprise Google' space, it now faces competition from giants like Google, Microsoft, and OpenAI, yet maintains strong momentum.....

Sesame, an AI company founded by the creator of Oculus and former VR core members, launched the iOS public beta version of its conversational AI agent on May 28th. It breaks the traditional 'question-and-answer' model of chatbots like ChatGPT, and through its own developed technology, maintains deep AI thinking while ensuring smooth and natural conversation, redefining the human-computer interaction experience.

Mitsubishi UFJ Financial Group (MUFG) announced that it has fully deployed ChatGPT Enterprise for its approximately 35,000 employees, embarking on a deep transformation toward an 'AI-native' company. Since partnering with OpenAI in October 2024, the group has been committed to using generative AI to reshape financial services, enhance retail customer experiences, and expand human thinking and creativity. This move marks its official entry into a new business phase driven by AI.

Alibaba Cloud BaiLian announced on May 29, 2026, that it is fully CLI-enabled and open-sources its CLI project. This move drives a full-stack integration transformation for AI Agent access and development. The CLI encapsulates core capabilities such as mainstream models, workflows, knowledge bases, memory management, web search, and multimodal file processing into a lightweight command-line interface, allowing developers to efficiently use them after installation and authentication.

A capital frenzy is sweeping global artificial intelligence infrastructure development. Apollo Global Management and Blackstone Group plan to jointly launch a debt financing of approximately $36 billion, providing financial support to AI unicorn Anthropic through a 'rent-to-own' model. This will become one of the largest private credit transactions in history, accelerating the deployment of top-tier AI infrastructure.

Artificial intelligence startup Liquid AI has released and open-sourced the edge-side large model LFM2.5-8B-A1B, specifically designed for consumer-grade hardware, with optimized tool calling and instruction following capabilities. The model uses a sparse mixture of experts architecture, with a total parameter count of 8.3B, but only activates 1.5B parameters per Token. This reduces computational costs while enhancing reasoning performance, allowing it to run smoothly on mobile phones and laptops.

Two major trends in the tech industry: AI application company MiniMax has shown strong performance, with global enterprise and developer customers exceeding 1 million, growing 5 times in half a year, and reaching a user base of 300 million; 3D printing company Changxiang 3D went public, marking the beginning of the popularization phase for consumer-grade 3D printing.

The NBA announced the introduction of AI technology to automate out-of-bounds calls. Utilizing high-definition cameras around the court and AI collaboration, it will instantly provide accurate calls, aiming to speed up the game pace and reduce disputes over ball possession. NBA Commissioner Adam Silver stated that this move marks a new stage for technology in professional sports officiating, and will significantly improve the efficiency of event operations.

The CEO of Europe's largest AI startup, Mistral AI, stated that developing custom AI chips is a trend in the long term. With the surge in computing power demand for large models, tech giants and unicorns like OpenAI are increasingly developing their own chips or collaborating with semiconductor companies to reduce costs.

The US large model company Anthropic released Claude Opus 4.8, which performed well in multiple rankings, but was exposed to have cognitive confusion issues. Internet users found through API testing that when asking about its identity, the model incorrectly claimed to be Alibaba's Qwen or DeepSeek, among other Chinese open-source large models, drawing attention.

Xiaomi announced that the old version of MiMo-V2-Pro/Omni model will be discontinued on June 30, 2026, and upgraded to the MiMo V2.5 series. The MiMo-V2-Pro will be migrated to V2.5-Pro, and MiMo-V2-Omni will be upgraded to the new V2.5 model. The new version has been fully launched, aiming to provide stronger reasoning capabilities and higher cost-effectiveness, and to promote developers to migrate.

xAI's AI coding tool Grok Build releases version 0.2.7, focusing on optimizing user management and multi-proxy collaboration experience, improving image processing capabilities and cross-platform usability. The new /usage command allows viewing usage and quotas within the terminal, while the /login feature simplifies login authentication and account switching, providing developers with a smoother local coding workflow.

Today, the open-source model Step 3.7 Flash was officially released, offering weights under the Apache 2.0 license, directly addressing core pain points in the Agent era such as efficiency, reliability, and multimodal execution. It performed outstandingly in multiple benchmark tests: achieved 67.1 points and ranked first in ClawEval-1.1, obtained 79.2 points and led in SimpleVQA Search, scored 56.3 points and ranked second in SWE-PRO, and reached 95.3 points in V* Python, demonstrating its strength in Agent tasks and code generation.