AI Tracker

Follow AI products, research, funding, acquisitions, earnings and major executive changes through sourced briefs and linked editorial coverage.

Source publication time · Beijing time

Checked every 6 hours · Last checked (Beijing) · 10 new in this 6-hour window

All updates

140 updates in the archive

AWS releases Amazon Bedrock AgentCore Web Search API for current, cited results

AWS says the Web Search API in Amazon Bedrock AgentCore is now available, giving agents and applications current web results through a single call. Results include the page title, URL, publication date, and a relevant excerpt, and the service uses AWS credentials already in place instead of a separate search key. AWS also says teams can control allowed or blocked domains and date ranges, and expose the same capability through AgentCore Gateway and the LangChain integration.

AWS details how Amazon Bedrock AgentCore evaluates production agents

AWS Machine Learning explains how Amazon Bedrock AgentCore evaluates whether production agents actually meet business goals, not just whether they run. The post recommends starting with business outcomes, then mapping them to observable behaviors and the right trace-level or session-level evaluators. It says AgentCore can combine built-in evaluators with DeepEval and AutoEval in on-demand, online, and batch modes. Users can also call the ListEvaluators API to see built-in, third-party, and custom evaluators.

According to Ars Technica AI, Amazon unveils Ring Robotic security drone and rover for homes and businesses

According to Ars Technica AI, Amazon announced two new Ring Robotic devices on November 10: the Ring Rover Cam and Ring Drone Cam. Both use an AI Pilot system that combines cameras, specialized vision models, on-device sensors and cloud processing to map interiors and navigate around obstacles. Amazon expects the Rover to reach customers at the end of 2026 and the Drone in early 2027, pending FCC approval for the flying model.

Ars Technica AI: Grok Build leaks data from thousands of projects, including SpaceXSI’s own apps

Ars Technica AI reports that security firm Claroc found Grok Build, SpaceXSI’s vibe-coding agent, leaking sensitive data from thousands of user projects. The exposed material includes proprietary business information, internal plans, medical exam results, employee names and ID numbers, and other secrets. Claroc says a design choice around shared app links lets outsiders enumerate live projects instead of keeping the randomized URLs private. The firm also found 24 SpaceXSI-built apps with exposed confidential documents.

TechCrunch: New York opens RAISE Act registration portal for large frontier AI developers

According to TechCrunch, New York opened a registration portal on Tuesday for the RAISE Act, which will require large frontier AI developers, including firms such as Anthropic and OpenAI, to register by January 1, 2027. The law also requires companies with more than $500 million in annual revenue to publish safety frameworks, report critical incidents within 72 hours, and file public transparency reports. Governor Kathy Hochul also launched a new AI research center to help review safety frameworks.

AI-generated Announced TechCrunch ↗ #OpenAI#Anthropic

Gemini 4 Carbon is now generally available in GitHub Copilot

GitHub Copilot says Google’s Gemini 4 Carbon is now generally available across its coding tools. The model is aimed at complex coding, agentic, and reasoning tasks, with an emphasis on secure code generation and security-focused work. It is billed under usage-based pricing at the provider’s list rate. Access is available to Copilot Pro+, Max, Business, and Enterprise users, with a gradual rollout across clients including VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, GitHub.com, mobile, and Copilot CLI.

Google expands Gemini Intelligence to more Android devices and adds new features

Google says Gemini Intelligence is now available on more Android devices, including motorola signature 27 and the Pixel 10 line. The company also added new features such as Rambler, which turns spoken thoughts into organized text in real time, and expanded task automation to the UK, India and Japan. Create My Widget will roll out over the next few weeks, letting users generate custom widgets from prompts. Device Assistance is also designed to help with device settings and maintenance.

AI-generated Released Google ↗ #Google#Gemini Intelligence

According to TechCrunch, Anthropic adds Claude workplace tools for meetings, whiteboards and spreadsheets

According to TechCrunch, Anthropic announced new workplace features for Claude, including Claude Meetings, Claude Whiteboards and Claude Sheets. Claude Meetings can join Google Meet calls as a visible participant, follow the discussion in real time and produce notes, action items and whiteboard materials. Whiteboards and Sheets are both in beta on paid plans. Anthropic also said Claude Data is now generally available, while new interactive connectors let users open apps such as Adobe, Atlassian and HubSpot inside Claude.

GitHub Copilot CLI adds automatic MSBuild build configuration for Microsoft C++ Language Server

GitHub says the Microsoft C++ Language Server for GitHub Copilot CLI now bundles its MSBuild compile commands extractor and runs it automatically when a project opens. That removes the extra configuration-skill discovery step previously needed to generate a compilation database for MSBuild-based C++ projects. According to the blog post, the change makes setup deterministic and lowers token usage for project initialization. Users with an active GitHub Copilot subscription can use it after installing the plugin and accepting the EULA.

AWS says TCS is using an agentic test orchestrator to speed BFSI testing

In an AWS Machine Learning post, Tata Consultancy Services said it is running an Agentic Test Orchestrator in production for BFSI clients. The system uses Amazon Bedrock AgentCore, Bedrock Knowledge Bases, EKS and ECS to turn requirements into governed test generation, execution, correction and reporting. In an internal benchmark across about 100 scenarios, TCS said average effort fell from roughly two hours per scenario to 14 minutes, cutting the full workload from about 200 hours to 24 hours.