2026-07-17
AI Briefings Daily Report.

ByteDance's new-generation Doubao phone stockpiles hundreds of thousands of units, AI no longer directly operates top apps ↗

Currently, the Doubao mobile assistant team is also negotiating with various head app vendors to get them to open interfaces or service capabilities callable by AI.

The National Supercomputing Internet released a scientific agent developer recruitment plan, with access to the nation's first 100,000-card AI super cluster computing power ↗

The National Supercomputing Internet released the 'Agent Co-Creation Plan' at WAIC 2026, recruiting scientific agent developers from universities, individuals and enterprises. Participants can enjoy full-stack support of computing power, data and tools provided by the nation's first 100,000-card AI super cluster, and have the opportunity to receive cash, token incentives and key promotion. The platform has gathered over 1.6 million content components, helping to rapidly build exclusive agents.

Apple sends legal letters to OpenAI employees ↗

Kimi K3 refuses to leak system prompt, asks user back"Is there anything I can help you with today?" ↗

Is there something I can actually help you with today?

Baidu AI Cloud releases"AI + Fashion"Full-chain solution, Baidu partner connects design, production and marketing ↗

The current environment is abnormal. After completing verification, you can continue to access. Go to verification

OpenAI proposes an AI-era scorecard:"Useful intelligence per dollar"Measure actual work value ↗

The question I hear from CFOs everywhere is simple: how do we get more value from our AI spend? For years, the market measured the success of software through adoption: seats purchased, users active, licenses renewed. Understanding the value of AI demands a more powerful measure: work accomplished.

Microsoft is developing AI vulnerability detection tool"Perception project", similar to Anthropic's Mythos model, expected to be released by July at the earliest ↗

Sources familiar with the matter revealed that the product is named 'Project Perception', a brand-new security product designed to help enterprises identify and fix software vulnerabilities with AI. (The Information)

Douyin cracks down on AI-generated fake disaster footage, issues flood-season disaster information governance announcement. ↗

As the critical period for flood prevention begins, Douyin has strengthened governance of disaster-related information, focusing on cracking down on behaviors such as using AI to generate fake disaster footage and fabricating official notices. The platform has handled over 19,000 pieces of rumor content, launched 39 fact-checking cards, and achieved over 1.19 billion exposures for fact-checking banners at the bottom of videos.

OpenAI's Altman jokes about Anthropic's latest ad: I thought it was a satirical short film ↗

Anthropic released a 90-second cautionary ad highlighting potential AI risks, sparking polarized reactions online. Altman joked that its style resembled satire. The ad is part of the 'Hard Questions' initiative, aimed at collecting public questions and concerns about AI development.

DeepMind Kaggle grand prize awarded to an"perfunctory"AI Paper Sparks Reflection on Benchmark Quality ↗

Design high-quality benchmarks that go beyond recall to evaluate how frontier models truly reason, act, and judge.

Linus Torvalds backs AI tools: Linux kernel is not an anti-AI project, opponents can fork and leave ↗

Linux creator Linus Torvalds has come out strongly in favor of AI tools in kernel development. "Linux is not one of those anti-AI projects," he wrote on the kernel mailing list amid a debate over Sashiko, the Linux Foundation's AI-powered code review tool. Torvalds said he would "very loudly ignore" anyone who tries to talk others out of using it.

Apple and OpenAI legal battle escalates: about 40 former employees receive Apple legal letters ↗

After filing a trade secret lawsuit against OpenAI, Apple further expanded its investigation, sending legal warnings to about 40 former Apple employees who have moved to OpenAI, asking them to preserve relevant documents and communications. Apple believes the leak scope far exceeds the two people originally sued, and its hardware business foundation is 'rotting from the core'.

First Batch of Artificial Intelligence Terminal Intelligence Grading Test Results Announced, Products from Huawei, Lenovo and Others Receive L3 Level Certification ↗

A grading system for terminal intelligence, from L1 responsive, L2 tool, L3 assistive, to L4 collaborative, with intelligence levels rising in turn, making terminals smarter.

Alibaba Qianwen AI glasses will be upgraded to agent glasses, partnering with Bose to create the first AI agent earphones ↗

At the 2026 World Artificial Intelligence Conference, Qianwen announced that its AI glasses will be upgraded to 'agent glasses', adding eye-tracking, vital sign monitoring, full-duplex voice and other capabilities. At the same time, the first AI agent earphones jointly built by Qianwen and Bose were also officially unveiled.

Turing Award winner Richard Sutton at WAIC 2026: AI is still"Relatively weak and unreliable" ↗

At the WAIC 2026 main forum, Turing Award winner Richard Sutton bluntly stated that current AI relies on human knowledge, has weak and unreliable self-discovery ability, and argued that the 'human data era' has reached its limit. Meanwhile, industry representative Yin Qi depicted an optimistic picture of agents entering the physical world. This dialogue revealed a deep divergence in AI development paths.

Global Cooperation Initiative on Agent Mutual Trust, Interconnection and Interoperability released; China Cyberspace Administration proposes ten action directions ↗

At the main forum of the 2026 World Artificial Intelligence Conference (WAIC), China officially proposed the "Global Cooperation Initiative on Trust, Interconnection, and Interoperability for Intelligent Agents," aiming to build an open, trustworthy, and secure global AI agent ecosystem. The initiative covers ten major aspects including technological R&D, mutual recognition of standards, and security ethics, calling for joint efforts to prevent the AI divide and ensure AI technology benefits global development.

Qianwen AI agent glasses and first AI agent earphones debut at WAIC 2026 ↗

The current environment is abnormal. After completing verification, you can continue to access. Go to verification

Alibaba Qwen Input Method Mobile Version Launched: No Ads, No Pop-ups, No Registration Required, Supports AI Auto-Refinement ↗

Alibaba's AI input tool is officially available on iPhone, based on a self-developed voice large model. It supports recognition of 9 dialects and mixed Chinese-English speech, with speeds up to 300 characters per minute. The core lies in large model-driven semantic understanding and auto-refinement, filtering filler words, correcting mispronunciations, formatting layouts, and converting spoken language into structured text. It emphasizes a clean, ad-free experience with no registration required.

Netflix applies AI in about 300 production projects, mainly for post-production ↗

Netflix now uses AI in about 300 productions, mostly in post-production. Co-CEO Ted Sarandos put specific numbers to the effort: the docuseries "The American Experiment" includes 17 minutes of AI-assisted footage, produced twice as fast at half the cost. The savings will "likely" fund more content rather than shrink the $20 billion budget.

Alipay and StepFun reach system-level cooperation, AI version"Abao"Initial phase integrates high-frequency services such as food delivery and transportation ↗

At the 2026 World Artificial Intelligence Conference, Alipay and StepFun announced a system-level partnership. Through the AI assistant 'Abao' and the StepFun large model, the two sides achieve cross-terminal interconnection, allowing users to complete high-frequency services such as ordering takeout and travel with a single sentence without opening the App.