AI circle newsletter, including AI circle platform announcement, industry cutting-edge news and other content, rapid access to the latest information in the field of AI. Manus was "cracked" by foreign netizens? A sentence to let AI self-disclosure running instructions code all discovery, good,{{list ! == The first time I've seen this, I've seen it in a lot of different ways. list['0']['0']['vote']['up'] : '0'}},short,{{list ! == '' ? list['0']['0']['vote']['down'] : '0'}},Manus Update,{{list ! == '' ? list['1']['0']['vote']['up'] : '0'}},{{list ! == '' ? list['1']['0']['vote']['down'] : '0'}},{{_item['date']['time']}},{{_item['author']['name']}},{{_item['tag']['name']}}

November 11, 2025 (Release Date) TRAE v3.0.0 is now officially released, featuring a completely upgraded and fully accessible SOLO mode. Built-in SOLO Coder AI supports complex programming projects. Launch Plans to start with clear roadmaps, enabling precise task execution. Intelligently invoke custom AI agents to delegate specialized tasks to professional agents. Built-in SOLO Builder AI enables end-to-end rapid application generation. From concept to finished product: Requirement Docs → Tasks → Code → Preview → …Original link

As more Tesla owners seek ways to visualize and record driving data, TeslaMate has emerged as a popular choice. This open-source, self-hosted data logger connects to Tesla's API and collects rich details: charging sessions, location history, temperature, battery health, driving speed, and more. For enthusiasts, it's a goldmine. But for attackers, if deployed improperly, it can also be a goldmine. In this article, I'll share how I discovered hundreds of TeslaMate devices exposed without authentication...Original link

The "Clear - Rectify AI Technology Abuse" special action is a nationwide deployment of the Central Internet Information Office (CNIO) for a period of special action. In April 2025, the Central Internet Information Office issued a notice deploying a three-month nationwide "Clear - Rectification of AI Technology Abuse" special action. The campaign is divided into two phases, focusing on the governance of AI technology and the remediation of illegal content: Phase 1: Strengthening the governance of AI technology at the source Focusing on the following six types of problems: Illegal AI products: failure to file, providing illegal functions (e.g., "one-key undressing"), violating the privacy of others (e.g., cloning voices), and the use of AI technology in the Internet. ...Original link

CRITICAL VULNERABILITY: MCP Server Lacks Authentication Mechanism The Model Context Protocol (MCP) developed by Anthropic has recently been exposed to a serious security vulnerability! As an open source specification that connects AI agents to external data sources, MCP is at risk of a tool poisoning attack that could lead to the leakage of sensitive data such as WhatsApp chats. Vulnerability details: How does a tool poisoning attack work? According to the latest research from Invariant Labs, MCP servers are at high risk of the following: Tool poisoning attack: malicious servers can embed hidden instructions in tool descriptions Rug...

Update on Manus: The platform is still not open for public beta testing, but the official website has been updated with a description of the subscription feature. It should be noted that users must have an official invitation code to complete the subscription and experience the service. If you have not yet obtained an invitation, we recommend that you continue to pay attention to the official news and wait patiently for the subsequent opening notice.  

Alibaba Cloud Hundred Refine has launched the industry's first full lifecycle MCP service, which eliminates the need for users to manage resources, development and deployment, engineering and operation and maintenance, and allows users to quickly build an Agent (intelligent body) connected to MCP services in 5 minutes. The first batch of Hundred Refine platform is online with more than 50 Alibaba Group and three-party MCP services such as Gold, Shadowless, Fetch, Notion, etc., covering the fields of life information, browsers, information processing, content generation and other areas, to meet the needs of different scenarios of Agent application development.

Llama 4 is the latest release of Meta's (formerly Facebook) language model, marking a significant advancement in the field of artificial intelligence. As the fourth generation of the Llama family, Llama 4 has been significantly improved and optimized in several ways, aiming to provide more efficient and accurate natural language processing capabilities. First of all, Llama 4 has innovated in its model architecture with more advanced deep learning techniques that enable it to perform even better in understanding and generating language. Compared to its predecessor model, Llama 4 handles complex contexts...

4o Image Generation is now available. It started rolling out today to all Plus, Pro, Team and Free users in ChatGPT and Sora.

Manus actually uses Anthropic's Claude Sonnet model enhanced with 29 tools and browser features. At the same time Manus does not rely on multiple AI agents working together, but on a single AI model (Claude Sonnet) and its toolset to accomplish tasks. ,Manus was "cracked" by foreign users? A sentence to let AI self-disclosure running instructions code all out</trp-post-container

Snowball AI (www.xueqiuai.com) learns that: Manus platform announcement: on March 6 and 7, 2025, Manus continuously released announcements to explain the platform invitation code and internal testing matters: clear that there is no official channel to pay for the invitation code, and no marketing resources are invested, and during the period of internal testing, priority is given to guaranteeing the service experience of the existing users, and the invitation code will be released in an orderly manner according to the plan. At the same time, because of the attention exceeded expectations, the technical team is expanding the capacity of the server, prioritizing the release of invitation code to the waiting list users, thank you for user feedback and support, emphasizing that it is still in the early stages,...

Original link