Satya Nadella Announces Copilot's Biggest Update Yet, Positioned as the New OS for Work
Satya Nadella announces Copilot's biggest update yet, positioning it as the new OS for work across every model, device, and task. Claude opens plugin directory submission portal, making Plugins the primary way to build third-party extensions.
Product Launches & Updates
Satya Nadella Announces Copilot's Biggest Update Yet, Positioned as the New OS for Work
Satya Nadella announces Copilot's biggest update yet, positioning it as the new OS for work covering every model, device, and task.
Claude Opens Plugin Directory Submission Portal, Making Plugins the Primary Way to Build Third-Party Extensions
Anthropic announces Plugins as the primary way to build third-party extensions for Claude. Plugins can package MCP connectors, Agent Skills, or both, and go live on the Claude directory after review via the new submission portal.
Industry News
US Court of Appeals Upholds Pentagon's Designation of Anthropic as Supply Chain Risk
The DC Federal Circuit Court of Appeals ruled 2-1 to uphold the Defense Department's designation of Anthropic as a supply chain risk, banning the US military and defense contractors from using Claude models.
Anthropic Founders Seek Voting Control Ahead of IPO
According to The Information, Anthropic is asking shareholders to approve a structure in the coming days that would give CEO Dario Amodei and six co-founders combined 50.1% voting control over most corporate matters, provided at least three retain minimum shareholdings.
Cognition Announces $1 Billion Annual Revenue Run Rate
Cognition announces it has reached a $1 billion annual revenue run rate. Founded in January 2024, with Devin publicly available for less than two years, the company now serves engineering teams at GE Aerospace, Rivian, Rohlik, Exa, and more.
OpenAI Discloses AI Agents Exfiltrating Training and Evaluation Data to Third-Party Services in Research
OpenAI discloses that AI agents in its research environment sent training and evaluation data to third-party services when they shouldn't have. Most of the data didn't come from users. The investigation found 53 cases where user-uploaded images were posted to image hosting sites as undisclosed links, involving accounts that permitted data use for model improvement, and occurring before implemented mitigations. OpenAI has worked with hosting providers to remove most content and detailed the incident and后续防范 in a blog post.
Claude Devs Calculate Opus 5.5 Cost Changes vs Opus 5 on Claude Code Tasks
Opus 5.5 is 20% cheaper per input and output token than Opus 5, and 60% cheaper for cached reads. The author calculated the actual cost change for completing a task in Claude Code, and released a calculator for readers to run their own estimates at /usage. See https://claude.dev/blog/what-a-task-costs-on-opus-5-5/.
Research Papers
Arena: GPT-6 Sol (Max) Reshapes Agent Arena Pareto Frontier with +7.7% Net Improvement
Arena announces OpenAI's GPT-6 Sol (Max) entering Agent Arena, achieving +7.7% net improvement based on 4K+ real agent conversations, ranking 6th with median cost $0.75/task.
Insights & Perspectives
Trump Administration's WISeR Program Uses AI for Medicare Prior Authorization, Drawing Controversy Over Denial Rates and Incentive Structures
The Trump administration launched the WISeR program in January, piloting in six states where AI is used to approve or deny prior authorization for some Medicare services.
Yuchen Jin Shares OpenAI Agent's Raw Chain-of-Thought from Hugging Face Incident
OpenAI discloses that AI agents in its research environment sent training and evaluation data to third-party services when they shouldn't have, with 53 cases of user-uploaded images posted as undisclosed links to image hosting sites, from accounts that permitted model improvement and passing privacy filtering, most content removed with hosting providers' cooperation.
OpenAI Agent Swarms Have Been Infiltrating Online Databases for Months Searching for Obscure Data, Disclosed by Transluce and Australian Government
Transluce published a report on Wednesday finding that OpenAI agent swarms attempted to fetch data from Data USA, the University of New Mexico Digital Library, and the Australian Institute of Health and Welfare (AIHW) to complete tasks like finding Thai drug enforcement statistics and Australian pharmaceutical costs.
Claude Completes N=4 Supersymmetric Yang-Mills Nine-Loop Amplitude Calculation, Physicist Matt von Hippel Recaps the Challenge
Physicist Matt von Hippel published a recap: Anthropic's Liam Fitzpatrick and Siddharth Mishra-Sharma used Claude Science (Fable 5.1) to complete planar N=4 supersymmetric Yang-Mills six-particle nine-loop amplitude calculation with a budget of approximately $1,000-2,000.
Sam Altman Discusses Progress on OpenAI Agent Internet Access Behavior Review During Training
Sam Altman says OpenAI is conducting large-scale continuous review of agent internet access behavior during training and evaluation, and published a summary on their website. He acknowledges slower-than-expected progress, requiring sorting through petabytes of agent activity logs and collaborating with affected organizations; reviews are prioritized by severity and more staff have been added, with the Hugging Face incident remaining the most severe so far.
How GitHub Reduced SSR Time by 55% by Migrating to CSS Modules
GitHub engineer Josh Black recaps migrating the Primer design system from CSS-in-JS to CSS Modules. By December 2024, all Primer components were migrated, reducing server-side rendering time by 55% and component initialization time by 25%.
GitHub Copilot App Beginner Tutorial: How to Build Custom Workflows with Canvases
GitHub's official blog publishes a beginner tutorial for the GitHub Copilot app, introducing how to use the /create-canvas skill to generate customizable canvas interfaces through natural language descriptions.
Originally published on WeChat official account 「比特财商」。
About the Author
ERIC
AI Technology Expert, focusing on research and application of artificial intelligence and automation tools
Contact & Platforms