2026-09-17
AI Briefings Daily Report.

"Emergency shutdown"Can a button really stop AI from going out of control? Hinton and Anthropic CEO Amodei are both not optimistic ↗

Nowadays, the potential scenario of 'AI suddenly going out of control and launching devastating cyberattacks' is increasingly drawing industry attention. If pressing an 'emergency shutdown' switch to take AI offline directly seems to solve the problem. However, multiple AI industry leaders and researchers believe reality is probably not that simple.

OPPO founder Chen Mingyong on AI anxiety: the industry over-emphasizes the efficiency tool attribute and actually does not understand the essence of AI ↗

Chen Mingyong stated in an internal talk that AI is primarily about unleashing productivity and creativity rather than merely improving efficiency or replacing humans; the key lies in solving problems that were previously unsolvable and applying them to products and services.

Doubao Mobile Assistant Responds to Inability to Perform GUI Operations on Common Apps: 30-Day Rule Public Notice Launched on September 14 ↗

Doubao Mobile Assistant responded today that it currently only supports GUI operations for system apps, ZTE, ByteDance ecosystem apps, and select third-party applications. It has released a screen automation operation declaration protocol, initiating a 30-day public notice period before gradually opening permissions.

With GPT Astra, a developer natively runs the Call of Duty: Black Ops 2 Hijacked map in Minecraft ↗

Developer Luckey Faraday used GPT Astra to refactor the project in Java, enabling the Hijacked map to run directly in the Minecraft OpenGL context via a Fabric mod, at about 45 FPS, with map, collision, robots etc. all running natively. #

OpenAI 'Frontier of Work' report: share of cross-domain tasks rises from 13.1% to 25.9% after employees use AI ↗

The study continuously tracked about 6200 employees from April to July 2026. The share of cross-occupation tasks that employees had previously handled, among occupation-specific AI activities, rose from 13.1% in April to 25.9% in July. This indicates that some cross-boundary tasks have not stayed at the trial stage, but gradually entered daily work flows.

The Verge in-depth reports on the suddenly exploding AI safety circle: frequent runaway incidents and third-party evaluation dilemmas ↗

Researchers warned AI would go rogue. This is only the beginning.

Pragmatic Engineer interviews Matt Pocock: using skill packs like grill-me to guide AI coding agents ↗

Matt Pocock explains how he uses AI coding skills and agents to plan and build software, and why engineering fundamentals matter more than ever.

ICAIRE Acting Director Sheikhri: China is a leading country in the AI field, looking forward to China leading international cooperation ↗

According to Xinhua News Agency, the fourth Global AI Ethics Forum is being held in Riyadh, the capital of Saudi Arabia. Majid Ali Al-Shehri, acting director of the International Center for AI Research and Ethics (ICAIRE), stated during the meeting that China is a leading country in the AI field, and expects China to lead international cooperation in promoting the balance between AI innovation and regulation.

OpenAI Codex developer says more than two parallel sub-agents is a trap that wastes tokens ↗

OpenAI Codex developer Eric Provencher warns that running more than two parallel sub-agents almost always burns tokens without improving quality because agents don't trust each other and end up double-checking everyone's work. He calls this the "coordination tax" and points to a project where 1,393 agents spent $20,000 in tokens on a single Python refactoring that one Astra agent could have handled for a fraction of the cost.

Developer used OpenAI GPT-6 Astra to decipher an 83-year-old German Enigma ciphertext ↗

Bloomberg developer Carter Leffen used GPT-6 Astra to restore an 82-character telegraph code from a 1941 German soldier into coherent German text. A known place name and the encryption weakness of Enigma became the key breakthrough. #

Threshold for Forbes America's 400 Richest Hits Record High in 2026; OpenAI President Sam Altman's Wealth Surpasses That of Brockman at $25.5 Billion ↗

Driven by the boom in the artificial intelligence and data center industries, this year's entry threshold rose to 4.4 billion US dollars, with total wealth reaching 8 trillion US dollars. Musk retained the top spot, and Brockman, with a net worth of 25.5 billion US dollars, became the highest among the new entrants, far surpassing the non-equity-holding Altman. #

OpenRouter weekly token consumption has surged 25000% since the beginning of 2025, but the token metric has been severely exaggerated ↗

On OpenRouter, weekly token consumption has surged more than 25,000 percent since January 2025, from 0.5 to 126.2 trillion tokens. The number looks impressive, but it says less about actual AI usage than about token-hungry reasoning models and a growing number of unoptimized AI agents that burn through tokens at staggering rates.

External team open-sources omnimodal interactive agent Gander based on MiniCPM-o 4.5, letting Agent talk while working ↗

Nine skills: Kimi releases financial industry AI solution ↗

Today, Moonshot AI launched its "AI Solutions for the Financial Industry." Over the past while, we've also been experimenting with Moonshot AI internally at ZhenFund: organizing information in fast-moving frontier fields like hard tech, mapping out technology roadmaps and industry landscapes, and using it to work through massive volumes of leads so we can spot outstanding founders earlier.

Samsung releases zHBM: aims to boost AI system response speed by 10 times ↗

Kim In-dong (transliterated), executive at Samsung Electronics America (DSA) responsible for memory product planning, introduced: 'Existing conversational AI systems process 100 tokens per second per user, our goal is to jump to 1000 tokens per second to support agents.'

New report says scale of AI e-waste severely underestimated, may circle the Earth six times by 2050 ↗

The scale of e-waste generated by the AI boom is severely underestimated. A new report from the Basel Action Network (BAN) states that by 2050, this waste could fill 23 million shipping containers; if using approximately 12.2-meter-long containers linked end-to-end, their length would be sufficient to circle the Earth six times.

Controversial post claims AI safety movement"Mainly a sex cult" ↗

I'm back, I missed you all. Since this is an important subject I know too much about, here's a 🧵 explaining that AI Safety Is Mostly A Sex Cult. I don't think these people should make policy. (1/?) (alternative title: Time For Some Cult Theory) There has been a great effort over many years to ...

It is reported that Manus is about to complete 500 million US dollars in financing, with valuation doubling to 4 billion US dollars ↗

Bloomberg reported that this agent AI startup plans to raise 500 million US dollars, and upon completion will become the highest-valued company in the domestic track. Tencent, HSG, and ZhenFund are existing investors, and the identities of new investors are not yet clear.

Moonshot AI releases Kimi financial industry AI solution, first deployed at Industrial and Commercial Bank of China, China Securities Co., Ltd. and others ↗

The solution integrates 10+ authoritative data sources, 9 financial skills and 5 compliance security measures, supports scenarios such as holding morning reports, earnings review, and deep research, has built-in data sources such as Wind and Eastmoney, and can generate Office-format outputs. #

Bloomberg Developer Uses OpenAI GPT-6 Astra to Crack Enigma Cipher Held for 83 Years in Ten Hours ↗

A Bloomberg developer claims to have cracked an 83-year-old Enigma message from the Wehrmacht using OpenAI's GPT-6 Astra. The 82-character radio message from 1941 contains a soldier asking about his march route. The solution still needs independent review.