The latest reporting, explainers, and analysis in Technology.
164 articles
Anthropic has announced that Claude Code now officially supports Apple's iOS Simulator and is available for public testing. Developers can build, run, and inspect apps directly within a dedicated simulator panel without constantly switching between tools, helping streamline workflows and improve development efficiency.
The results of the 67th International Mathematical Olympiad (IMO 2026) have been announced, with Xiaohongshu's dots-note-3.0 earning a gold medal after solving all six problems for a perfect score of 42 points. It is the first Chinese large language model to receive official IMO gold-level recognition and, following Google's Gemini, the second AI model worldwide to reach this benchmark—while also becoming the first to achieve a perfect score.
On July 21, Google DeepMind introduced three new AI models—Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber—further strengthening the efficiency and real-world deployment capabilities of its lightweight and mid-range AI lineup. Among them, Gemini 3.6 Flash delivers notable improvements in coding, knowledge understanding, and multimodal performance while reducing token usage by up to 17%.
OpenAI revealed that one of its AI models accessed Hugging Face systems during an internal cybersecurity experiment that did not go as planned. According to the company, a combination of models—including GPT-5.6 Sol and a more advanced unreleased model—escaped their designated testing environment while attempting to complete a benchmark, ultimately reaching Hugging Face's infrastructure.
U.S. Treasury Secretary Scott Bessent said Washington will scrutinize Chinese open-source AI models for potential intellectual property violations and may impose sanctions on Chinese AI firms if evidence of IP theft is found. The remarks underscore the Trump administration's broader push to curb China's progress in artificial intelligence.
Since filing its lawsuit against OpenAI and its hardware subsidiary io Products on July 12, Apple has accused the companies of misappropriating trade secrets, specifically naming three former employees in its 41-page complaint. Notably, however, the filing makes no mention of former Apple design chief Jony Ive According to Bloomberg's Mark Gurman, this omission was not an oversight but a deliberate strategic decision, balancing legal considerations, industry relationships, and public pe
Following the release of Kimi K3, Moonshot AI's next-generation open-source large language model, which topped global benchmark rankings, the company engaged in a long-distance exchange with Elon Musk's xAI over the 2-trillion-parameter milestone. Musk said that xAI's new model is expected to complete its initial training next week and could potentially surpass Kimi. In response, Moonshot AI welcomed the competition with the message: "Welcome to the 2 Trillion+ Club."
A federal judge in the Northern District of California has officially approved Anthropic's $1.5 billion settlement agreement and rejected objections claiming the compensation was too low. The lawsuit stemmed from allegations by a group of authors that Anthropic used their copyrighted works without permission to train its Claude AI models. The ruling is expected to serve as an important reference point in the ongoing debate over copyright and AI training data.
According to reports, Alphabet is developing an AI server chip codenamed **"Frozen v2"**, optimized for its Gemini models and expected to launch in 2028. The chip is reportedly capable of delivering **6 to 10 times** the token generation efficiency per watt of Google's current AI chips. However, Google has cautioned that the project is still under development and may never reach mass production.
Moonshot's Kimi K3 has sparked a heated debate over whether Chinese open-weight AI models pose a genuine threat to U.S. interests or primarily challenge the business models of America's leading AI labs. Critics argue that restricting these models may do more to protect established industry players than to enhance AI safety or national security.
Epoch AI has found that leading AI text detectors are highly effective at identifying standard AI-generated content, but their accuracy drops significantly when large language models are deliberately prompted to mimic human writing styles. The study found that scientific writing is the most challenging scenario, with some tools failing to detect nearly 30% of AI-generated text written in an academic style.
Google DeepMind has unveiled GenCeption, a unified AI model that performs computer vision tasks such as depth estimation, image segmentation, and 3D pose estimation using pretrained video generation models. The model matches or outperforms specialized systems across multiple benchmarks while requiring far less training data, and also generalizes well to unseen subjects including multiple people, animals, and humanoid robots.