AI News
Signal over noise. Daily.
一線 AI Lab
21 articlesAgent SDK reference - TypeScript
Complete API reference for the TypeScript Agent SDK, including all functions, types, and interfaces. npm install @anthropic-ai/claude-agent-sdk. The SDK bundles a native Claude Code binary as an optional dependency; set pathToClaudeCodeExecutable if your package manager skips it, and note bun build --compile can't resolve the bundled CLI.
Agent SDK reference - Python
Complete API reference for the Python Agent SDK. pip install claude-agent-sdk. The SDK provides two ways to interact: query() creates a new session per call for single exchanges, while ClaudeSDKClient reuses the same session for multiple exchanges in the same context.
Use Claude Code features in the SDK
Load project instructions, skills, hooks, and other Claude Code features into your SDK agents. When you omit settingSources, query() reads the same filesystem settings as the CLI: user/project/local settings, CLAUDE.md files, and .claude/ skills, agents, and commands. Pass settingSources: [] to run with only what you configure programmatically.
Agent Skills in the SDK
Extend Claude with specialized capabilities Claude autonomously invokes when relevant. Skills are packaged as SKILL.md files. In the SDK they are defined as filesystem artifacts in .claude/skills/, loaded from locations governed by settingSources (TS) or setting_sources (Python), and automatically discovered.
Work with sessions
How sessions persist agent conversation history, and when to use continue, resume, and fork to return to a prior run. A session contains your prompt, every tool call, every result, and every response, written to disk automatically. Sessions persist the conversation, not the filesystem; use file checkpointing to snapshot and revert file changes.
Intercept and control agent behavior with hooks
Hooks are callback functions that run your code in response to agent events like a tool being called or a session starting. Use them to block dangerous operations, log and audit every tool call, transform inputs and outputs, require human approval for sensitive actions, and track session lifecycle.
Stream responses in real-time
Get real-time responses from the Agent SDK as text and tool calls stream in. By default the SDK yields complete AssistantMessage objects; enable partial message streaming by setting include_partial_messages (Python) or includePartialMessages (TypeScript) to true to receive incremental StreamEvent updates as tokens are generated.
Plugins in the SDK
Load custom plugins to extend Claude Code with skills, agents, hooks, and MCP servers through the Agent SDK. Plugins are packages of extensions that can be shared across projects and loaded programmatically from local directories. The commands/ directory is a legacy format; use skills/ instead.
Streaming Input
The Claude Agent SDK supports two input modes: Streaming Input Mode (default and recommended), a persistent interactive session, and Single Message Input, one-shot queries that use session state and resuming. Streaming mode lets the agent run as a long-lived process handling input, interruptions, permission requests, and session management.
Slash Commands in the SDK
Slash commands control Claude Code sessions with special commands that start with /. Only commands that work without an interactive terminal are dispatchable through the SDK; the system/init message lists the ones available in your session. Access available commands from the system initialization message when your session starts.
Our views on AI policy and political advocacy
Our approach to AI policy and political advocacy, transparency, support for thoughtful regulation and AI safety, and that no outside political group speaks on the company's behalf.
How we used Gemini to build Google I/O 2026
Learn how Googlers used AI to produce Google I/O 2026.
Confidential Draft S1 Sec
Anthropic has confidentially submitted a draft S-1 registration statement to the Securities and Exchange Commission.
Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains
A blog post by JetBrains on Hugging Face introducing Mellum2, a 12B mixture-of-experts model.
Series H
Anthropic has raised $65 billion in Series H funding led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital.
Beyond LLMs: Why Scalable Enterprise AI Adoption Depends on Agent Logic
A blog post by IBM Research on Hugging Face on why scalable enterprise AI adoption depends on agent logic.
Building the infrastructure for the Intelligence Age in Michigan
OpenAI breaks ground on a 1GW data center project in Michigan as part of Stargate, building AI infrastructure to expand access, create jobs, and support communities.
Claude Opus 4.8
Claude Opus 4.8 is an upgrade to the Opus class of models, with stronger performance across coding, agentic tasks, and professional work, and the consistency to handle long-running work.
OpenAI frontier models and Codex are now available on AWS
OpenAI frontier models and Codex are now generally available on AWS, giving enterprises a new path to build with OpenAI through the AWS environments, controls, and procurement workflows they already use.
OpenClaw Collaborates with NVIDIA for Stronger Agent Skill Security
Every ClawHub skill now ships with a Skill Card documenting what the skill does and where it came from, and is scanned by SkillSpector for hidden instructions and other agentic risks.
Welcome NVIDIA Cosmos 3: The First Open Omni-model for Physical AI Reasoning and Action
A blog post by NVIDIA on Hugging Face introducing Cosmos 3, the first open omni-model for physical AI reasoning and action.
媒體
36 articlesAlphabet plans to raise $80B to pay for AI buildout
"The company is experiencing strong demand for its AI solutions and services from enterprises and consumers, at levels that are exceeding the company's available supply," Alphabet said in its statement.
Can Content Creators Get Rich Off A.I. Slop Like Tung Tung Tung Sahur?
Brainrot videos are specifically designed to confuse us. They can also be a financial bonanza for their creators.
AI costs how much? GitHub Copilot users react to new usage-based pricing system.
As GitHub's new usage-based billing for Copilot goes into effect, many users report sticker shock at how quickly their previous "normal" usage burns through their limited monthly AI credits. For some, it reportedly took less than a day to use up a month's quota.
Nvidia chases $200B CPU market with AI agent PCs from Microsoft, Dell, and HP
If Nvidia has cracked a way to bring AI agents easily, safely, and usefully to the masses, it could — and should — be big.
The Joy of Hate-Watching A.I. Slop
A.I. slop is reconstituting a shared cultural base line.
Hackers duped Meta AI support chatbot to steal celebrity Instagram accounts
Meta's AI support chatbot proved unusually helpful to hackers, who simply asked the bot to change accounts' associated email addresses while masking their locations with a VPN. The exploit let them take over and flip notable accounts before Meta issued an emergency patch on May 29.
Bernie Sanders: A.I. Belongs to the People, Not to Billionaires
All Americans should have a stake in the future of this technology.
How Box Created 13 New Types of Jobs Because of A.I.
Box, a Silicon Valley software maker, expects to have more employees, not fewer, as it hires A.I. architects, A.I. solutions managers and other new A.I.-related positions.
Florida sues OpenAI, Sam Altman, in first-of-its-kind lawsuit over violent incidents
The lawsuit partially revolves around a shooting at Florida State University last year, and ChatGPT's alleged role in the incident.
This could be Windows' M1 moment — but expect it to cost a ton
Nvidia's move into consumer laptop chips with RTX Spark could be Windows' chance to blow us away with a new generation of capable Arm-based chips, much like Apple's M1 in 2020 — but the launch is expected to come at a steep price.
Gemini's new AI agent is about as good as Google's demo
Google's new "24/7" AI agent, Gemini Spark, can be shockingly good at doing multi-step tasks on your behalf — but it may not be worth the financial cost and potential privacy tradeoffs. Google says it's "always under your direction" and checks with you before major actions.
Tribeca Festival 25th Anniversary: An Interview With Robert De Niro, Jane Rosenthal, Rebecca Glashow
Robert De Niro and Jane Rosenthal, who founded the event, and Rebecca Glashow, the new leader of Tribeca Enterprises, reflected on the festival's legacy and ambitions.
Meta's own AI was exploited to hijack Instagram accounts
Meta's AI support chatbot helped hackers hijack Instagram accounts by switching the associated email and resetting the password. The issue, now patched, surfaced around the same time Barack Obama's White House Instagram account was hacked and posted Iranian propaganda.
Florida sues OpenAI, Sam Altman after multiple ChatGPT-linked murders
Florida became the first state to sue OpenAI over ChatGPT's allegedly dangerous design, with AG James Uthmeier accusing the company of prioritizing profits over safety. The suit follows two violent events in which suspects allegedly used ChatGPT to assist in planning.
Anthropic Files to Go Public, Setting Stage for Huge I.P.O.
The AI company, which is racing OpenAI to the stock market, has seen explosive growth over the last year thanks largely to technology that can automatically write computer code.
Florida Sues OpenAI Over Chatbot Safety Concerns
The state became the first to sue the ChatGPT maker over claims that its technology posed a risk to children and that the company had failed to warn the public of dangers.
California's Public Universities Went All in on A.I. Now They're Tearing Themselves Apart.
California's public universities spent $16.9 million on A.I. during a financial crisis, and the result has been chaos.
Water access is now a risk factor in SpaceX's IPO
The company says it needs "significant" water resources to cool its data centers, and that access to abundant, affordable water is a challenge.
From 15 hours to one minute: How AI/ML is speeding up GM's development
GM chief product officer Sterling Anderson describes how AI/ML is accelerating the automaker into what he calls the third epoch of engineering and design, collapsing tasks that once took 15 hours down to about a minute.
Anthropic Confidentially Files for What Could Be the Largest IPO Ever
The AI giant behind Claude submitted paperwork on Monday that would take it public, just a couple of weeks after SpaceX's splashy IPO announcement.
Allegedly trashing Airbnbs to test robots puts startup in legal trouble
A San Francisco robotics startup, The Bot Company, is being sued by an Airbnb host who claims its "robotic prototype testing" caused extensive damage to his home. Sean Donovan is seeking more than $12,000 in damages after finding bundles of wires and a 6-foot-tall robot throughout the house.
Germany Has Lost What It Did Best
The economy is in a terrible mess.
Anthropic has officially filed to go public
Anthropic filed to kick off its IPO process with the SEC on Monday, setting the stage for a massive offering. As of its fundraise last week the company is called the world's most valuable startup, with a post-money valuation of $965 billion, topping OpenAI's $852 billion.
Anthropic files to go public
Anthropic, now an AI powerhouse that has landed top-tier enterprise customers, was once considered an underdog in the emerging world of large language models.
This AI weather startup is out-forecasting government agencies
WindBorne benefits from its unique combination of model-building and data collection, with about 400 balloons in flight gathering sensor readings at any given time, launched from 15 sites around the globe.
DuckDuckGo makes its 'no-AI' search engine easier to access as its traffic booms
Alternative search engine DuckDuckGo launches 'no AI' web extensions for Chrome and Firefox users.
Microsoft to unveil new AI models and Windows improvements at Build
Microsoft heads to a smaller, more intimate Build conference in San Francisco to win back developers amid low trust in Windows and GitHub. Sources say we'll hear about new AI models in Windows, a new reasoning model from Microsoft AI, and a Copilot "super app."
AI is blowing up music. How should the Grammys handle it?
Recording Academy CEO Harvey Mason Jr. says AI is now "omnipresent" in music production, with every recent session involving AI. The legendary producer discusses what tools musicians are using and how the Grammys should handle the shift.
Is A.I. Replacing Tech Workers or Providing an Excuse for Job Cuts?
Tech industry layoffs are accelerating, and executives have been quick to say it's because their companies are doing more with artificial intelligence, even when there may be more to it.
China Aims A.I. at Predicting Who Could Pose a Political Risk
New research examines how a Chinese company struggled to develop its predictive surveillance technology while U.S. restrictions were in place.
Strava blames zero-code AI apps and scrapers as it tightens API access
Strava is restricting API access to clamp down on AI scraping, now requiring developers to pay a flat $11.99/month subscription. The company blames "zero-code AI tools" that hammer its APIs, citing a 448% year-to-date jump in developer applications and degraded performance.
Chinese Military Sought Nvidia Chips for Years, Report Says
An analysis of six years of procurement records suggests that the People's Liberation Army has openly tried to acquire restricted U.S. technology.
Nvidia Has a Plan to Put Its Chips in Personal Computers
The world's most valuable company is chasing Intel and Apple as it tries to bring A.I. agents to laptops and desktops.
Intel: Our upcoming AI chip will be cheaper, run cooler than Nvidia, AMD options
Intel plans to ship an AI chip by year's end that uses cheaper memory and cooling than Nvidia and AMD. Its new "Crescent Island" GPU targets inference tasks rather than training, where Nvidia dominates, as the chipmaker seeks to capitalize on a turnaround.
Norse Atlantic Airways Offers Dirt-Cheap Tickets. There's a Catch
Dozens of people have complained to the Federal Trade Commission about Norse Atlantic Airways' tech-first customer service operation. Some said they lost thousands of dollars.
An OpenAI model solved a famous math problem that stumped humans for 80 years
In mid-May, OpenAI announced that an internal model disproved the Erdős unit distance conjecture, a discrete-geometry problem unsolved for 80 years. Fields medalist Tim Gowers called it a milestone, and Daniel Litt called it the first AI-produced result he finds exciting in itself.