According to CoinWorld, Anthropic has launched an agent identity security mechanism for its team collaboration product, Claude Tag, aiming to completely block backdoors for AI data leaks. This mechanism equips each channel with an independent permission package, directly assigning dedicated accounts to AIs to prevent unauthorized access in multi-person collaborations. The new security architecture replaces the traditional borrowing of personal credentials with dynamic policy configuration files, enabling AIs to function as independent team-level identities. Security sandbox expert Kenton Varda criticized the model, arguing that directly assigning dedicated accounts to AIs is unsuitable for large-scale tasks and undermines human accountability mechanisms. He pointed out that agents cannot be legally and administratively held accountable; all operational permissions must originate from a real, living person. Varda advocates for a capability-based security model to ensure that AI actions are traceable to the specific initiator.

According to CoinWorld, OpenAI's ChatGPT subscription and quota system experienced a widespread anomaly, causing confusion regarding usage quotas and model permissions for users at different levels. Some Plus subscribers suddenly displayed a 20x limit, gaining access to the higher quotas previously reserved for Pro subscribers. Simultaneously, some Pro subscribers paying $200 per month found their usage quotas abnormally reset to 75%, or their quota limits increased by four times. The system malfunction also affected model usage permissions; many Pro subscribers reported that while their account subscription status displayed Pro, their actual performance had been downgraded to Plus level, and the GPT-5.5 Pro flagship inference model was unavailable in the model selection list. To date, OpenAI's official status page only indicates the issue may occur when users are using Codex access tokens, stating that mitigation measures have been implemented. The official stance has not been publicly acknowledged or explained regarding the quota calculation and account level synchronization anomalies experienced by ChatGPT subscribers. The community generally believes that a faulty backend fee calculation and synchronization logic is the underlying cause of the rapid depletion of quotas and abnormal rate limiting.

According to CoinWorld, Notion has announced a public beta of its external agents platform as part of its business and enterprise initiatives. This platform supports integrating Anthropic's Claude and AI programming tool Cursor into team collaboration dashboards. Users can @ agents or directly assign tasks within the Notion workspace, allowing AI to automatically collaborate on code writing, data analysis, and content generation. The newly beta Claude agent runs on the Anthropic Managed Agent (CMA) architecture. Users do not need to configure API keys and are paid for per run using Notion credits. After task assignment, the model can autonomously read documents and execute multi-step planning. However, it's important to note that the CMA architecture operates in a stateful manner; Notion's Zero Data Residue (ZDR) commitment does not apply to the Claude agent, and session data will remain on the server. Notion has also opened an external agent interface, allowing users to bind their local Cursor agent to the Notion workspace for zero Notion credit consumption and automated development.

Anubis Labs warmly congratulates VEST on its upcoming global launch, marking a new stage in its ecosystem development. As an ecosystem partner, Anubis Labs will continue to provide VEST with ecosystem resources, technological collaboration, and global development support, jointly promoting the continuous innovation and expansion of the digital capital, asset tokenization, and Web3 finance ecosystem. In the future, both parties will engage in deeper cooperation in the following areas: Digital capital infrastructure construction; RWA (Real-World Asset) digitization and tokenization; Web3 finance ecosystem collaboration; Global market expansion; Ecosystem resource interconnection and cooperation. Leveraging Anubis Labs' ever-growing ecosystem network, VEST will further connect global developers, partners, and community users, continuously expanding the application boundaries of digital assets and on-chain finance. The global launch is a new starting point and a significant step towards a global ecosystem. Congratulations to VEST on embarking on its new global journey! 🚀🌐

According to CoinWorld, GLM-5.2, a model developed by Zhipu AI, has received official ARC-AGI benchmark validation, demonstrating performance comparable to OpenAI's lightweight GPT-5.5. Data shows that GLM-5.2 achieved an accuracy of 22.8% on the ARC-AGI-2 benchmark, with an average run cost of $0.25 per task. On the slightly less challenging ARC-AGI-1 benchmark, GLM-5.2 achieved an accuracy of 77.0%, with a run cost of $0.19 per task. ARC Prize, the benchmarking organization, stated that testing of GLM-5.2 on the more challenging ARC-AGI-3 benchmark is currently suspended due to limitations in inference throughput; the team plans to release specific test results only after a reliable evaluation is possible.

According to Bijie.com, social influence analysis platform xhunt released version v0.3.4 today, officially expanding from Web3 social analysis to both AI and Web3 domains. This version introduces an AI social influence system for the first time, launching AI KOL influence rankings, AI KOL capability analysis, and AI real-time trending topic tracking. xhunt's AI KOL capability analysis model evaluates creator value based on content performance, influence spread, interaction quality, and topic participation. In addition, the Web3 trending list adds a Chinese-language KOL interaction ranking and an 8-hour trending tweet ranking, and upgrades the multimedia section to aggregate popular videos and long-form articles. xhunt v0.3.4 is now officially available for download and updates.

According to CoinWorld, AI startup Mirendil has announced the completion of a $200 million seed funding round, valuing the company at $1 billion. The round was led by Andreessen Horowitz and Kleiner Perkins, with participation from NVIDIA. Mirendil aims to develop recursive, self-improving models capable of autonomous optimization and iteration, assisting scientists in developing specialized AI models for fields such as medicine and materials science. Founders Behnam Neyshabur and Harsh Mehta met while working at Google in 2019 and joined Anthropic together in late 2024, subsequently leaving after the release of Claude Opus 4.5 in December 2025. Mirendil currently has an office in downtown San Francisco and approximately 20 technical staff. The team plans to release models and products in the coming months to gather user feedback.

According to CryptoWorld, Anthropic has launched a new security architecture for its Claude Tag team collaboration feature, completely eliminating account sharing and assigning each AI a dedicated, independent account. This new architecture changes the old model where AIs used employee accounts to perform tasks, ensuring that AIs use dedicated accounts when posting on Slack, submitting code on GitHub, or querying data repositories. Enterprise administrators can centrally configure the tools and code repositories that AIs can access and finely adjust access levels across different channels to ensure that sensitive conversations are not leaked. The system will also support instant authorization in the future, minimizing the risk of permission abuse.

According to data from the State Administration for Market Regulation, from January to April 2026, the number of enterprises in my country's "8+9" (8 emerging industries and 9 future industries) sectors maintained steady growth, with a total of 373,000 new enterprises established, further releasing innovation vitality. Among the specific sub-sectors, the generative artificial intelligence field saw 34,000 new enterprises established, a year-on-year increase of 24.9%; the humanoid robot field saw 78,000 new enterprises established, a year-on-year increase of 13.3%; and the quantum information field saw over 7,000 new enterprises established, with a growth rate of 8.9%.

According to CryptoWorld, Anthropic, OpenAI, and Stripe have jointly launched the Intercept consortium, announcing a $500 million investment to develop new antiviral drugs and promote air purification technology. The initiative is co-led by Nan Ransohoff, Stripe's Head of Public Goods, with contributors including Flu Lab, employees of the Jane Street exchange, and Bill Gates' philanthropic arm. The Intercept program will support the development of broad-spectrum vaccines and drugs for respiratory viruses such as the common cold, COVID-19, and influenza, with follow-up development handled by major pharmaceutical companies. Simultaneously, it plans to promote existing air purification technologies and build a network of potential buyers, including JPMorgan Chase, Mastercard, and Meta. Respiratory infections impose a significant burden on global public health; estimates suggest that the COVID-19 pandemic claimed approximately 14.9 million lives in 2020 and 2021, and research by the UK's Institute for Health Economics indicates that respiratory infections cost UK businesses approximately £44 billion annually.

According to CoinWorld, Unsloth AI announced that it has compressed the size of Zhipu AI's 753B parameter large model GLM-5.2 by more than 80% using dynamic quantization technology, and released a GGUF format version that supports local deployment on Mac. Through dynamic 1-bit and 2-bit quantization, the original 1.51 TB model can be reduced to 217 GB (1-bit variant) to 239 GB (2-bit variant), allowing ordinary developers and small and medium-sized enterprises to deploy and run it locally offline using only a single Mac Studio. The quantized version achieved a smooth speed of 21.6 tokens/s on a Mac Studio M3 Ultra (256 GB unified memory) device, while retaining 76% to 82% of the original model's accuracy. Currently, the GLM-5.2 GGUF weights are available for download on the Hugging Face platform, and users can load and run them directly through llama.cpp or Unsloth Studio.

According to data from Crunchbase, as of June 2026, the total valuation of the world's top 30 unicorn companies reached $3.9 trillion, with AI startups dominating, accounting for approximately 60% of the total valuation. Among these companies, Anthropic tops the list with a valuation of $965 billion, followed by OpenAI with a valuation of $852 billion. Other AI companies on the list include Databricks, Figure, Safe Superintelligence, Anysphere, Scale AI, and Cognition. In the fintech, commerce, and mobility sectors, Stripe, Revolut, Checkout.com, Ramp, ByteDance, Ant Group, and Shein are among the top ten.

According to CoinWorld, Gate.AI recently completed an upgrade to its end-to-end large model management platform, launching a one-stop large model routing service for enterprises and developers. The platform now integrates over 200 mainstream large models globally, supporting both OpenAI and Anthropic protocols. Enterprises can access different model resources through a single API, achieving unified access and management, and reducing development, operation, and migration costs. The platform has built a multi-level unified management system, combined with zero data retention and data processing protocols to strengthen privacy protection, helping enterprises achieve efficient, standardized, and transparent operation of AI resources. In the future, Gate.AI will continue to deepen its expertise in model access, intelligent routing, enterprise governance, and application innovation, creating an open AI ecosystem across the entire value chain and providing long-term support for the intelligent upgrade of global enterprises.

According to CoinWorld, JPMorgan Chase has raised its year-end 2026 target for the S&P 500 from 7600 to 7800, citing upward revisions to corporate earnings expectations and increased capital expenditure on artificial intelligence. JPMorgan Chase points out that low-quality growth stocks, speculative growth stocks, and second- and third-tier AI concept stocks face extreme crowding risks, and capital outflows could trigger a correction. JPMorgan Chase recommends a barbell strategy, holding high-quality growth stocks and stocks directly benefiting from AI, while also allocating to low-volatility, high-quality stocks.

According to CoinWorld, Sakana AI has released Fugu, a multi-agent collaborative system that enables dynamic orchestration of multiple models through a single API. In several authoritative industry benchmark tests across academia, inference, and programming, the top-of-the-line version, Fugu Ultra, outperformed Anthropic's flagship model, Fable 5, and Mythos Preview. Fugu Ultra is now available on OpenRouter, supporting tool calls and multimodal input. Model input costs $5 per million tokens, output costs $30, and cache reads cost $0.50.

According to Bijie.com, Thailand's exports are expected to reach a record $366.8 billion this year, primarily driven by global investment in artificial intelligence, which is fueling demand for Thai electronics. Nantapong Chiralerspong, Director of the Office of Trade Policy and Strategy of Thailand, stated that total exports are projected to grow by 8% this year compared to 2025. Electronics currently account for nearly one-third of Thailand's exports, and this sector is expected to become a major driver of growth as global technology companies increase their investment in AI infrastructure. This trend has also boosted Thai tech stocks, making its benchmark index the best-performing market in Southeast Asia this year.

According to Bijie.com, Xiuyu Li, a graduate of the Turing Class at Renmin University, has announced that he has joined Stepfun, a domestic large-model vendor, as a large-model post-training researcher. During his time at university, Xiuyu Li focused on research in large-model alignment and reinforcement learning (RL). His academic achievements include the training-free inference-time alignment method ETS, published as a co-first author at ICML 2026, and the paper DART on the conflict between inference and tool invocation in agent reinforcement learning.

According to Bijie.com, Shi Kaiwen, former VP of Product at Lark, has confirmed his departure to start his own business, focusing on edge AI agents. The core team for this project is largely complete, with core co-founding members including renowned professors from Shanghai Jiao Tong University, and some members from ByteDance and Shanghai Jiao Tong University. In terms of funding, Sequoia China and Jinqiu Fund have participated in the early-stage investment, and overall fundraising is progressing smoothly. This represents another significant early-stage bet in the edge AI agent field in recent years. During his time at Lark, Shi Kaiwen was responsible for product direction for a long time, rising to the position of VP of Product. He was deeply involved in Lark's transformation into an AI collaboration tool and attended the Boao Forum for Asia in March of this year as Lark VP. Currently, the company's name and specific product form have not been disclosed.

According to CoinWorld, Google's main model, Gemini 3.5 Flash, natively supports PC control, unlocking enterprise-level AI agent automation. This feature is directly integrated into the main model as a built-in tool, eliminating the need for developers to use a dedicated Gemini 2.5 PC to control the model. After integration, developers and enterprise users can control devices via the Gemini API or the Google Cloud Gemini Enterprise AI Agent Platform, simplifying AI agent development architecture. The built-in tool receives screenshots for visual perception and step-by-step reasoning, outputting operation commands such as mouse clicks and keyboard input to automate tasks like software testing and data collection. To prevent the risk of prompt word injection, Google has conducted targeted adversarial training on the model and provides protective measures such as manual verification and task circuit breaking. Currently, browserbase provides an online hosted demo environment, and Google has also open-sourced the reference implementation code on GitHub.

According to Bijie.com, David George, a SpaceX investor and partner at Andreessen Horowitz LLC, stated that the "rapid reusability" of the new Starship rocket is crucial to realizing Elon Musk's vision of expanding AI computing power through orbital data centers. He pointed out that once Starship is operational, extending AI capabilities into space will become inevitable, as the capacity of AI data centers on Earth is increasingly limited. George proposed a new perspective on these future satellites: "aircraft-sized GPU racks deployed in space." He stated that orbital data centers will exist in space as supplementary computing power to existing Earth-based facilities, and from an economic perspective, space infrastructure may ultimately be superior to ground-based facilities. He also noted that SpaceX has already demonstrated the technological capability to deploy such systems.