Newshot Article title: Morning Brief: 700 Number Segment Protection Service Pilot Approved, Claude Code Settings Files Leak Credentials, and More
Body:
700 Number Segment Protection Service Pilot Approved
On November 10, the Ministry of Industry and Information Technology announced that it has approved 10 enterprises to carry out pilots of number protection service business, and has allocated dedicated 700 number segment code number resources for the pilot enterprises' number protection service business. Previously, the Ministry of Industry and Information Technology issued the "Notice on Carrying Out the Pilot of Number Protection Service Business," planning the 15-digit 700 number segment as a dedicated code number for number protection service business, to address the problem of irregular use of mobile phone numbers and other numbers as intermediate numbers. The pilot enterprises will cooperate with basic telecommunications enterprises and internet platforms, giving priority to scenarios such as food delivery and ride-hailing; users can independently choose whether to enable the number protection service. After activation, riders, drivers, and others can communicate with users bidirectionally through intermediate numbers starting with 700, and during the call the real numbers of both parties are hidden from each other. Users can also learn the usage scenario and platform entity of the incoming call through caller ID display, pre-connection voice announcements, and other methods. The 10 pilot enterprises are: China Telecom Digital Intelligence Technology Co., Ltd., China Mobile Internet Co., Ltd., Alibaba Cloud Computing (Beijing) Co., Ltd., Tencent Cloud Computing (Beijing) Co., Ltd., Guangxi Dongxin Yitong Technology Co., Ltd., Beijing Lingdu Technology Co., Ltd., Beijing Zhibo Jiuzhou Information Technology Co., Ltd., Beijing Tianrun Rongtong Technology Co., Ltd., Beijing Weihu Technology Co., Ltd., and Xi'an Tiecheng Network Technology Co., Ltd. The Ministry of Industry and Information Technology stated that next it will, on the basis of the pilot, promote the coverage of number protection services to more enterprises and more scenarios. Source
Claude Code Settings Files Leak Credentials
On November 9, AI security company Apiiro discovered that Claude Code settings files are leaking sensitive information such as database passwords, API tokens, and root login credentials in public GitHub repositories. When users approve shell commands and select "Yes, and don't ask again," this authorization is written as a rule into the local .claude/settings.local.json file. The rule is sometimes recorded as a pattern, but in many cases the command itself is written verbatim, including the sensitive information embedded in it; Anthropic documentation states that this file will be added to the developer's global .gitignore, but this setting is not reflected at the repository level. If the repository is cloned to another machine, or if the folder later becomes a repository, the file may be mistakenly uploaded. Apiiro also stated that in many cases Claude Code will commit changes to this file on its own without user intervention.
In this instance, Apiiro found more than 100,000 such settings files uploaded to public GitHub repositories, of which about 1,100 files from 63 organizations contained secrets. Among these secrets, 215 were not identified by the two commonly used secret scanning tools gitleaks and TruffleHog; even after the files are deleted, the secrets may remain in forks of the original project. The issue was first raised in Claude Code bug reports as early as August 2025, and the related reports were all closed after 60 days of inactivity. Boris Cherny of Claude Code said on Hacker News that the team is studying how to prevent this issue, while stating that such settings files basically should not be committed to Git repositories, and that the situation appears to occur only when "the file is manually created and committed"; he recommended that users use auto mode and secret scanning tools, and the team is also considering building secret scanning into Claude Code in the future. Source
National Standard for Satellite Navigation Reference Station Network Security Released
On November 10, the State Administration for Market Regulation (Standardization Administration) announced that it has approved and released the mandatory national standard "Security Requirements for Satellite Navigation Positioning Reference Station Networks" (GB 48567—2026). The standard was proposed and administered by the Ministry of Natural Resources, and specifies security technical requirements from three aspects: facilities, data, and services. It defines the definitions and technical identification features of reference stations and reference station networks, and divides reference station network services into two categories: surveying and mapping reference services and other social services; unapproved reference station facilities must not be set up within military restricted zones or military administrative zones; reference stations must be equipped with isolation and protection facilities, and access to and operation of station equipment shall be subject to hierarchical permission management; data processing, service, and management systems must obtain a cybersecurity classified protection level 3 filing certificate; reference station network data shall be subject to hierarchical control of classified data and important data according to sensitivity, raw observation data must be transmitted encrypted through wired private networks or commercial cryptography technology, and important data services shall implement real-name user verification and access control; real-time services involving sensitive data must be equipped with security isolation and protection measures, and reference station networks providing services to society shall be periodically traced to the national unified surveying and mapping datum. According to information recorded by the National Standard Library, the standard was released on October 30, 2026, and will take effect on November 1, 2027. Source
INMO Recalls 4,701 AIR3 Glasses
On November 10, the State Administration for Market Regulation announced that Sichuan INMO Technology Co., Ltd. has reported a recall plan to the State Administration for Market Regulation and the Sichuan Provincial Administration for Market Regulation in accordance with the "Interim Provisions on the Administration of Consumer Product Recalls," recalling some INMO AIR3 smart glasses manufactured between May 2025 and June 2026, totaling 4,701 units, effective immediately. The reason for the recall is that the glasses may have abnormal power consumption calls when running some third-party applications not officially tested, and the left temple may overheat during prolonged use, posing a burn hazard. INMO Technology will handle the issue free of charge through an OTA firmware upgrade. Users need to update the system to version V3.18.109 through official channels and should reduce usage time before the update is complete; INMO Technology will notify relevant users through system push notifications, and users can also call the customer service hotline 4008878167 for inquiries, with service hours from 9:00 to 21:00. Source
8 AI-Related News Items
MiniMax Opens H3-Image Weights
On November 11, MiniMax published an article announcing the image generation and editing model H3-Image and opening the model weights. The model is based on H3 and supports text-to-image, reference image generation, and image editing, with up to 16 reference images as input; MiniMax states that it can generate 4096 × 4096 pixel images within 10 seconds.
The official model page explains that the model directly generates images of about 16 million pixels without super-resolution upscaling, and supports aspect ratios from 1:1 to 21:9; generation and editing share one set of weights, and after distillation it uses 6-step sampling, with a diffusion Transformer parameter count of 33.1 billion. The weights are released under the MiniMax-H3-Image community license agreement, and API, MiniMax Code, and Hailuo AI entry points are also provided; the official announcement states that during the first week of launch, image generation requests for 4K and above will be charged at 50% of the standard price. Source
Antigravity Introduces Layout Mode Switching Option
On November 9, Google introduced an interface layout for knowledge work for Antigravity. The Antigravity app can now switch between the "Knowledge Work" and "Software Engineering" layouts. Both run the same Antigravity agent and both can use frontier models, local files, plugins, skills, rules, custom agents, and MCP servers. The Knowledge Work layout no longer prompts for Git worktrees and hides the Git branch selector and the button to jump to the IDE; tool call groups, inline code diffs, and sub-agent cards in the conversation are changed to a one-line real-time progress prompt, and permission requests are summarized in plain language; the sidebar removes "Review" and "Terminal," focusing instead on documents, artifacts, links, uploaded content, and background tasks; developer options such as the terminal, workspace, and detailed permission rules are moved into advanced settings.
Users can switch in "Settings > General > Interface Layout," which takes effect immediately without restarting. Existing users still default to the Software Engineering layout; new users can choose roles such as software engineering, knowledge work, data and analytics, science, or design when first logging in, automatically matching the initial layout, example prompts, and recommended plugins. Source
Gemini 4 Carbon Expands Availability
On November 10, Google announced the expansion of Gemini 4 Carbon's availability. Paid developers and enterprises can call the API through channels such as Google AI Studio and Gemini for business, and Google AI Ultra subscribers can gradually use it in Antigravity, the Gemini app, and AI Mode in Search.
Google states that Carbon is aimed at long-horizon tasks such as multi-step programming, debugging, knowledge work, and scientific research. Antigravity is also previewing the /swarm feature, which can hand a single prompt to a multi-agent team, with agents respectively taking on design, coordination, review, and parallel execution work. Carbon's API promotional price is $2 per 1 million input tokens and $10 per output token; after the promotional period ends on November 30, 2026, these will be adjusted to $4 and $20 respectively. Source
Google DeepMind Releases Gemini Robotics 2.5
On November 10, Google DeepMind announced the Gemini Robotics 2.5 vision-language-action model, which has been gradually opened to robotics partners. The official announcement states that the model was evaluated on more than 200 tasks and, compared with previous baselines, improved task success rates by 30% to 100% and completion speed by 20% to 30%; the Apptronik Apollo 2 equipped with the model achieved a 96% success rate in a parts sorting trial on a Mercedes-Benz production line. The team also open-sourced the robot safety testing benchmark ASIMOV-Jailbreak. Source
Gemini Launches "Memory Summary" Feature
Morning Brief: 700 Number Range Protection Service Pilot Approved, Claude Code Settings File Leaks Credentials, and More
On November 10, Gemini launched "Memory Summary," which lists what Gemini knows about the user by category, including core information, career and work projects, travel and daily gear, media and entertainment, and fitness and lifestyle. Users can view, edit, and delete entries, and can also use natural language in the input box at the bottom to request adding or removing information; if a request is unclear, Gemini may ask follow-up questions. "Memory History" in the overflow menu lets users view when specific information was added and modified.
On the web, it can be accessed via "Settings > Personal Intelligence." The feature is currently available to Google AI subscribers who have enabled "Keep Activity" and memory features and use US English. Source
Meituan Open-Sources LongCat-2.5
On November 10, the Meituan technical team announced in a post that LongCat-2.5 has been open-sourced under the MIT license, supporting commercial inference deployment and secondary training without additional authorization. LongCat-2.5 is a native multimodal model for long-horizon agent tasks, with 1.6 trillion total parameters, about 48 billion parameters activated per token, and support for a 1 million token context; visual signals are represented as multi-level discrete tokens and processed with the same set of backbone parameters as text, rather than through an add-on vision module; more long-horizon trajectories were added during training, and real agent failure trajectories were converted into reproducible, verifiable training tasks; inference uses a W4A4 scheme, with expert weights and activations both quantized to 4 bit; long-sequence training and fine-tuning were adapted for domestic AI accelerators with less than 80 GB of memory per card, and Meituan says its FSDP configuration improves single-card throughput by 86% over a DDP baseline in long-sequence SFT tasks.
Meituan says the new model shows clear improvements in end-to-end delivery capabilities in complex codebase development, complete project construction, Visual Coding, and professional office scenarios. The model has been open-sourced on Hugging Face, and an API platform is also provided. Source
Ant Open-Sources Tactile Model LingBot-Touch
On November 10, Ant Lingbo Technology open-sourced the tactile foundation model LingBot-Touch. The model was pretrained on nearly 5,000 hours of real tactile data, covering three types of sensing principles: magnetic tactile, visuotactile, and piezoresistive, and can convert signals from different sensors into a unified tactile representation for use by different dexterous hands and action models; the model sets encoding entry points by sensing principle, maps contact points with different layouts into a unified fingertip coordinate space, and outputs unified tactile tokens, without being tied to Ant Lingbo's self-developed action models.
Ant Lingbo completed 26 real-machine tests on two dexterous hands, Sharpa Wave and XHand1, including turning book pages, twisting bottle caps, and identifying mahjong tiles. After integrating LingBot-Touch, the average task success rates of four action models—LingBot-VLA, π0.5, Diffusion Policy, and T-Rex—increased by about 14 to 26 percentage points compared with no tactile sensing; among them, the magnetic tactile sensor model used by XHand1 was not in the pretraining data, and the model parameters remained unchanged during testing. The model is available in two capacity versions, and all 26 tests were completed using the lightweight version LingBot-Touch-Tiny with about 5.3 million parameters, adding only about 0.13% to 0.16% to the overall parameter count after integration. The technical report, model, and code have been open-sourced on Hugging Face, GitHub, and ModelScope. Source
Claude Expands Artifacts with Whiteboards, Sheets, and More
On November 10, Anthropic previewed Claude Meetings and launched beta versions of Claude Sheets and Claude Whiteboards.
Claude Sheets and Claude Whiteboards are extensions of the Artifacts feature. Sheets supports creating and editing spreadsheets with a team, including multiple worksheets, formulas, formatting, and dropdown menus; Claude can directly modify cells and explain each change, and files can be exported as Excel or CSV. Whiteboards supports teams in organizing ideas, timelines, owners, and dates in real time, and can convert whiteboard content into documents or presentations. The upcoming Claude Meetings can add Claude to calls, take notes during discussions, and update team documents.
At the same time, Claude Dashboards was renamed Claude Data and officially launched, adding features still in beta such as authenticated metric queries and real-time sharing. Sheets, Whiteboards, and Data are available to all paid users on web, desktop, and mobile; they are enabled by default for Team plans, while Sheets and Whiteboards are disabled by default for Enterprise and must be adjusted by administrators. Source
Apple TV to Support Sign Language Interpretation
On November 10, Apple announced that it is adding sign language interpretation to the Apple TV app, planned to launch globally on November 13 with the film Being Heumann. The film is based on the real-life experiences of disability rights advocate Judy Heumann. F1 will also offer American Sign Language and British Sign Language versions, and the feature will expand to Way of the Warrior Kid and Greyhound within the year, with more Apple Original titles to follow. The sign language performances were designed and tested with participation from deaf people and native sign language users, and supported films will display a sign language icon; users can first enable the feature in the Apple TV app settings, then select the sign language button during playback, and can also search for terms such as "with ASL" to find related content. The feature supports iOS 27.2, iPadOS 27.2, macOS 27.2, tvOS 27.2, and visionOS 27.2 and later systems, as well as supported third-party devices running the latest version of the app. Source
Brief News Worth a Glance
- The Cybersecurity Threat and Vulnerability Information Sharing Platform of the Ministry of Industry and Information Technology, together with the first batch of 18 vulnerability collection platforms including Butian Vulnerability Response Platform and Vulbox, signed the "Self-Discipline Convention for Online Product Security Vulnerability Collection Platforms" at the 2026 Network and Data Security Industry Forum; the convention has 4 chapters and 18 articles, setting out agreements on platform filing, capacity building, user management, and the entire process of vulnerability collection, assessment, handling, and disclosure. Source
- GreyNoise said its honeypot in Southeast Asia was breached by the attack group GT10, and the OpenSSH and musl library interaction vulnerability "*sshole," disclosed on November 5, has already been exploited in the wild. The vulnerability can allow attackers to log in without authentication within seconds to high-privilege accounts on some Linux systems, with systems such as Alpine more severely affected; OpenSSH has released a fix, and Ars Technica recommends that users update as soon as possible. Source
- Apple released iOS and iPadOS 27.2.1, 26.7.3, and 18.7.12; macOS Golden Gate 27.2.1, macOS Tahoe 26.7.3, and macOS Sequoia 15.8.3; watchOS 27.2.1 and 26.6.1; and visionOS 27.2.1, fixing multiple security issues. Source
- Google began rolling out the November Pixel Drop in batches. The Pixel 10 series (excluding 10a) adds Gemini Intelligence, supporting organizing dictation content and generating widgets with natural language; long conversation summaries add support for German, French, Spanish, Italian, Hindi, and Traditional Chinese. Source
- Mozilla released Firefox 160, supporting an on-device real-time speech recognition interface for 40 languages, as well as MKV files containing FLAC, MP3, or PCM audio; picture-in-picture adds playback speed settings from 0.25x to 2x; WebGL on macOS switches to the Metal graphics interface. Source
Recent Updates from SSPAI
- We have set up a dedicated page to summarize and display the "Outstanding Author Awards" and their results. The "Outstanding Author Awards" is a mechanism implemented by the editorial team since 2023. After the end of each quarter, it recognizes existing authors who made outstanding contributions during that quarter, as well as emerging new authors with potential. Winners receive an exclusive badge and prizes. Go to view
- Sonos Play 15% off with a charging kit, and a CD player with a free transparent cover. Shop now
- Limited-time discounts on SSPAI membership, genuine software, and paid columns. Don't miss out
Articles You May Have Missed
- Community Express 165 | Transparent Badges and SSPAI Friends' October Shopping Lists
- Record Number of Registrations, the 11th "Mobile Application Innovation Contest" Concludes Successfully
- These Mice Are Worth a Look This Double 11
Download the SSPAI client and follow the SSPAI official account to unlock a new reading experience 📰
Practical, useful genuine software, presented to you by SSPAI 🚀