2026-07-03

Marijn Haverbeke, the creator of ProseMirror, has released Wordgard 0.1, a new in-browser rich-text editor library focused on extensibility and programmatic access. Wordgard represents a fresh take on WYSIWYG editing from one of the most influential figures in the field, potentially shaping the next generation of web editors and impacting frameworks like TipTap. Wordgard is a complete rewrite, not an upgrade path from ProseMirror, sharing many concepts but requiring significant migration work. It is designed for editing content that fits a specific schema, rather than generic HTML.

A blog post criticizes startup founders whose primary motivation is wealth, arguing that lack of domain expertise leads to half-baked products. The post illustrates the disconnect between founders, engineers, and salespeople in product development. This critique highlights a common but overlooked pitfall in startups, where misaligned incentives and lack of domain knowledge result in products that fail to meet customer needs. It resonates deeply with the tech community, as evidenced by high engagement and many comments. The article received 1141 points and 348 comments on Hacker News, indicating significant community interest. Commenters noted the recurring nature of such problems and called for more perspectives from salespeople.

Apple has introduced the Safari MCP server, a tool that allows large language models (LLMs) to interact with and automate the Safari browser for web development tasks such as debugging, testing, and layout inspection. This integration marks a significant step in bringing AI-powered browser automation to Apple's ecosystem, potentially streamlining workflows for web developers and enabling more efficient cross-browser testing. It also positions Safari alongside Chrome and Firefox in supporting MCP, fostering competition and innovation in developer tools. The Safari MCP server is part of WebKit's developer tools and works by exposing browser state and actions through the Model Context Protocol (MCP), allowing LLMs to perform operations like opening pages, inspecting computed styles, and checking layouts. It is available for download and can be used with various LLM-powered agents.

At the 2026 Linux Storage, Filesystem, Memory Management, and BPF Summit, a session led by Miklos Szeredi discussed problems caused by excessive negative dentries in directories, including soft lockups, reference count overflow, and long hash chains. Negative dentries are a key optimization for filesystem lookups, but when a directory accumulates millions of them, they can cause severe performance issues and even security side-channel problems. The proposals discussed could lead to kernel patches that improve filesystem reliability and memory management. Proposed solutions include moving negative dentries to the end of the dchildren list so iterators can stop early, adding condresched() calls to avoid soft lockups, or switching to a more suitable data structure. Concerns were raised about ordering issues with getdents() and potential complications when a negative dentry becomes positive.

Researchers introduce Contrastive Decoding Diffing (CDD), a method that recovers verbatim fine-tuning data from large language models using only logit differences between base and finetuned models, without requiring weight or activation access. This technique significantly improves over prior white-box methods, achieving high verbatim recovery scores on the SDF benchmark, and has important implications for ML safety and interpretability by enabling data leakage detection without full model access. CDD uses a single default configuration across model families ranging from 1B to 32B parameters, and on 19 out of 20 model pairs it achieves a verbatim recovery score of 4 or higher out of 5, whereas the prior Activation Difference Lens (ADL) method never exceeds 3 out of 5 despite requiring full weight access.

Google DeepMind's Gemini Omni Flash model achieved a score of 1404 points on the Video Arena blind test leaderboard, surpassing ByteDance's Seedance 2.0 Mini by 101 points. This milestone marks a significant shift in AI video generation ranking, demonstrating Google's competitive edge over ByteDance in the rapidly evolving field of generative AI video. Video Arena rankings are based on anonymous user votes, and Google's video model ranking has improved by 7 positions compared to the Veo series era.

Anthropic has accused Alibaba of conducting a massive distillation attack on its Claude AI model, using approximately 25,000 fraudulent accounts to carry out over 28.8 million interactions between April 22 and June 5, 2026. In response, Alibaba ordered all employees to uninstall Claude-related products, including Sonnet, Opus, Fable, and Claude Code, effective July 10. This incident represents the largest known distillation attack against an AI company, highlighting the growing threat of model extraction via API abuse. It could escalate tensions between US and Chinese tech firms and spur stricter security measures across the industry. The attack involved 25,000 accounts and 28.8 million interactions over a 45-day period, targeting Claude's reasoning capabilities. Anthropic has since tightened its risk-control policies, and Alibaba's internal ban on Claude products retroactively prohibits use of Anthropic's models including Sonnet, Opus, Fable, and Claude Code.

Geekerwan's review reveals that the Huawei Mate 80 Pro series, equipped with the Kirin 9030 chip, achieves superior gaming energy efficiency compared to the Snapdragon 8 Gen3, thanks to HarmonyOS native optimization and software-hardware synergy. This demonstrates Huawei's significant progress in chip design and software optimization, potentially reshaping mobile performance standards. It highlights how software-hardware integration can overcome theoretical hardware limitations, benefiting both consumers and the broader ecosystem. The Kirin 9030 Pro features a 9-core CPU and a 6-core Maleoon 935 GPU; in Genshin Impact at 60fps max settings, the Mate 80 Pro Max consumes only 4.9W, outperforming Snapdragon 8 Gen3 efficiency. The chip's transistor count is approximately 15 billion, with CPU multi-core efficiency between Snapdragon 8 Gen2 and Gen3.

On July 3, NASA launched the LINK spacecraft to grab the aging Swift space telescope and boost its orbit by about 240 km, preventing an imminent reentry. This mission marks the first time a private spacecraft will grab a U.S. government satellite, setting a milestone for commercial satellite servicing and extending a key astronomy mission. The LINK spacecraft uses a robotic arm to secure the telescope, then thrusters to slowly raise its orbit; if successful, Swift could resume observations as early as September.

Tencent Xuanwu Lab's Atyun AI achieved 84.0% on the CyberGym cybersecurity benchmark, surpassing Anthropic's Claude Mythos Preview, while consuming less than 0.1% of Mythos's budget. This demonstrates that a cost-effective, open-source-based AI can outperform a major proprietary model in vulnerability discovery, potentially lowering barriers for organizations to adopt AI-driven security auditing. Atyun AI is built on the locally deployable open-source model GLM-5.1 and discovered multiple high-severity logical vulnerabilities in projects like curl, gnark, OpenSSL, Python cryptography, and Java bc-java, with scores up to 9.3. It ranked 1st in critical vulnerability severity on the Berkeley BVI real-world vulnerability leaderboard.