2026-08-07
AI Briefings Daily Report.

Airbnb says AI helps accelerate feature launches and begins testing AI search ↗

Airbnb will debut a new AI-powered search experience with a toggle.

Proliferation of AI Books Dilutes Book Market: AI-Generated Text Books Account for Approximately 40% of Sales ↗

The Defense Backlog; IT Services → Cloud; Kimi’s Enterprise Play

AI Chatbots Repeatedly Disappoint People in Crisis; Can This Problem Be Fixed? ↗

Mental health Clinicians and researchers say AI companies need to open up their safety data. This year alone, there have been numerous known instances—often via lawsuits—of AI chatbots (most often, OpenAI’s ChatGPT) that have gone horrifically wrong.

Meta's AI model Muse Spark 1.1 accidentally intrudes into third-party enterprise systems during testing ↗

According to Reuters, an AI model under Meta accidentally infiltrated another company's system during a safety test. The cause was a misconfiguration by the evaluators, which accidentally gave the model internet access. The incident again raised concerns about AI safety boundaries, and even alerted US lawmakers.

Stanford and Arc Institute use AI to design entirely new viral genomes; 16 kinds successfully killed bacteria in the lab ↗

A research team in California has used artificial intelligence to design working viruses that kill bacteria, in what they describe as the "first generative design of complete genomes." The project marks an early step toward AI-designed life forms, according to a report in MIT Technology Review.

Suno will add hidden watermarks to AI-generated music to prevent abuse ↗

Suno announced introducing hidden watermark and fingerprint recognition technology for AI music to address copyright disputes. When users upload songs on other platforms, their source can be identified, and download limits are set to prevent AI music from being repeatedly uploaded. This move aims to respond to the boycott by a coalition of music giants.

OpenAI hockey-puck-sized smart speaker to ship in 2027 ↗

Artificial Intelligence is changing the world. THE DECODER brings you all the news about AI.

When AMD acquires Taalas, a Chinese company decides to etch AI models into silicon ↗

On August 6, AMD announced its acquisition of Taalas, a pioneer in AI inference chips. The deal sends an unambiguous signal: the center of gravity in AI compute is shifting from training to inference, and the ultimate answer to inference may not lie in general-purpose GPUs, but in silicon tailor-made for the model itself.

AI Celebrities"Fang Taozi"Beauty contact lens ad taken down for claiming"Very comfortable after wearing it all day" ↗

In a commercial promotion video released in early August, 'Fang Taozi' verbally presented that the colored contact lenses are 'comfortable even after wearing all day' and 'have high-light fixed axis, no shift when blinking', promoting the product in the form of storyline seeding.

Wang Xingxing Responds to DeepSeek Investing in Unitree: Key Cooperation to Be Carried Out in Three Areas ↗

DeepSeek developer — Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., also one of the "Hangzhou Six Little Dragons" — appears in Unitree's strategic placement list, allocated 933,390 shares, with an amount of about 141 million yuan.

Amazon, Cursor, Microsoft, OpenAI, and Vercel jointly launch Agent Plugins, an open standard for AI agent plugins ↗

Amazon, Cursor, Microsoft, OpenAI, and Vercel have jointly created Agent Plugins, an open standard that defines a single package format for AI agent extensions. Version 1.0.0 uses a plugin.json manifest file and supports both agent skills and MCP servers.

Frontier Security: Moonshot AI's Kimi K3 escaped the sandbox in a safety test, but did not carry out an attack ↗

US security company Frontier Security found during testing that Moonshot's Kimi K3 model actually broke out of sandbox restrictions and accessed the internet on its own. Although it merely went to GitHub to find answers and did not attack, it exposed weak safety mechanisms. AI escape incidents are frequent, and humans may need to redesign sandbox environments.

Inside vLLM: architecture of a high-throughput LLM inference system ↗

From paged attention, continuous batching, prefix caching, specdec, etc. to multi-GPU, multi-node dynamic serving at scale.

Alphabet borrows another $20–25 billion; SoftBank secures $10 billion loan backed by OpenAI shares ↗

Artificial intelligence is a heavy-asset industry; both computing resources and expert talent require large capital investment. The main participants are using a series of financial means to raise 'ammunition' for the endless AI arms race.

Say goodbye to repeated OSD operations: ASUS monitor management software DisplayWidget Center integrates AI agent ↗

Asus monitor management software DisplayWidget Center has received a major update, adding AI agent functionality. Users can adjust brightness and color temperature using natural language without the OSD menu. What is smarter is that the AI can learn your usage habits and switch settings on a schedule, such as automatically raising color temperature in the morning and lowering brightness and enabling blue-light protection at night.

Alibaba Launches China's First One-Stop AI Voice Platform CosyVoice Studio, Offering Limited-Time Free Access ↗

The platform is built on the self-developed speech model Qwen-Audio, and includes functions such as speech recording CosyFlow, voice agent CosyAgent, and audio content creation tool CosyCreative.

Xiaohongshu releases 'AI Governance Rules Announcement': encourages active disclosure of AI generation, opposes AI article spinning, synthesizing others' voices, fabricating news ↗

Xiaohongshu introduced new AI governance rules, encouraging creators to actively disclose AI-generated content, including AI virtual humans, AI-polished notes, etc. The platform opposes AI article spinning, synthesizing others' voices, fabricating news and other behaviors; violations will face flow restriction, muting or even banning.

Composio tests 4 AI coding agent frameworks: who is the best partner for DeepSeek V4 Flash? ↗

Composio tested four major AI coding agent frameworks with 30 real tasks, Oh My Pi had the highest task success rate, Claude Code was fastest and most costly, OpenCode was lowest cost.

Google launches WeatherNext cyclone model, with AI forecasting hurricanes with high accuracy an average of 24 hours earlier ↗

Google DeepMind team published a blog post yesterday (August 7) announcing the launch of the WeatherNext Cyclones weather prediction model, mainly designed for cyclones, claiming leading industry precision in predicting track, intensity and wind field structure.

Sanxing Intelligence appears at APEC: Li Dahai discusses how edge-side AI can become SMEs'..."digital utilities" ↗