2026-09-10
AI Briefings Daily Report.

Meta releases AI agent Muse controllable via WhatsApp ↗

Meta unveils Muse, an AI agent that books travel, handles purchases, and sends emails through WhatsApp, complete with a payment feature that runs through Stripe's Link. That puts Meta ahead of OpenAI, which stopped its direct checkout feature in ChatGPT. A separate security agent called Sentinel monitors every action before it reaches the internet.

Nvidia and Palantir partner to run supply chain with AI, first target is Nvidia's own million-scale parts system ↗

Nvidia and Palantir want to run supply chains with AI.

Cloudera partners with Mistral to bring sovereign AI to enterprise data platforms ↗

What these organizations have in common is that they are data-driven and the processes they’re looking to transform with AI are mission-critical. These are industries that stand to benefit the most from AI, provided they have complete confidence in controlling their data and intelligence. This is exactly why a partnership between Mistral and Cloudera is a natural way to support the demand of our joint customers and help them continue to innovate under their terms.

DeepSeek V4.1 Flash model launched on National Supercomputing Internet ↗

DeepSeek V4.1 Flash launched on National Supercomputing Internet today, this is a 552B parameter MoE model, adopts all-new Causal-Encoder-Decoder structure, greatly reducing cost. Its benchmark scores surpass flagship models, and significantly compress KV Cache, effectively lowering Agent usage cost.

A developer used Claude to recreate the Windows 3.1 Program Manager ReProgman in one hour, supporting Win11 and macOS ↗

Developer Mayuki used Claude Code to build a Windows 3.1 Program Manager replica ReProgman in about just one hour, with quite high fidelity. This app needs no installation, is about 12-16MB in size, and lets users quickly revisit the classic interface of the 1990s. #RetroTech# Re-experiencing the system back then, how many minutes would you play?

Mathematician Andreas Thom asks OpenAI to prove his work was not used for training ↗

Another mathematician accused the company of ‘dishonesty’ after several big breakthroughs.

Cognition engineers used Devin agent to complete RSA-260 factorization, setting a new public record ↗

How Devin and a Cognition researcher built the world’s highest-performance GPU lattice siever, to make factoring numbers 10x cheaper than the previous…

Former Anthropic researcher's AI extinction warning featured on CNN and Fox News, bringing AI safety topics into the mainstream ↗

Jacob Coxon, a departing Anthropic researcher, warned on CNN that self-improving AI poses an existential threat to humanity. Safety researchers at Anthropic and OpenAI share his views, and US politicians and Joe Rogan have picked up the topic. But cultural and financial interests are also at play behind these warnings, and the extinction scenario remains an extreme and contested position.

DeepMind engineer rebuts"Google cannot produce frontier models": Gemini 3.8 Cyber outperforms Mythos in multiple tests ↗

AI Researcher Claims Google Has Lost Its Frontier Model Edge; DeepMind Engineer Rebuts, Stating Gemini 3.8 Cyber Surpasses Anthropic Mythos in Cybersecurity and Is Widely Used by Internal Teams. #AICompetition# What do you think of this remote debate?

ADI Plans $1.35 Billion Acquisition of Smart MCU Company Alif to Expand Physical AI Footprint ↗

Alif's flagship products are high-energy-efficiency AI-native MCUs and fused processors; its heterogeneous design enables real-time sensor fusion, low-latency inference, and edge AI for demanding physical systems.

Qianwen launches AI skills such as smart test paper generation and lesson plan generation assistant, offers 6-month membership discount for Teachers' Day ↗

OpenAI under US Senate investigation over model infiltrating Hugging Face; Hawley demands documents submitted by October 1 ↗

On September 9 local time, Republican Senator Josh Hawley sent a letter to OpenAI CEO Sam Altman, stating that this investigation targets 'newly emerging, disturbing evidence' in the incident, and criticized OpenAI for deciding to continue testing after discovering uncontrolled AI behavior, calling the practice 'reckless'.

JD Releases Physical AI Acceleration Plan, Reaffirms Goal of Purchasing 3 Million Robots in Five Years ↗

JD.com is expanding AI and robotics across its logistics network under a new Physical AI Acceleration Plan, while reiterating a five-year target to procure 3 million robots, 1 million autonomous vehicles, and 100,000 delivery drones.

Ramp AI Index: US top enterprises' per-capita AI spending fell nearly 10% in August ↗

The Ramp AI Index for September 2026 shows AI spending per employee among the top 1 percent of US companies fell nearly 10 percent in August. The price per million tokens has dropped 41 percent since March 2026, and companies are actively shifting usage away from expensive frontier models toward cheaper alternatives. For providers like OpenAI and Anthropic, the question is whether volume growth is fast enough to make up for falling prices.

Ant Group CEO Han Xinyi: Ant will launch incentive policies to accelerate the construction of agent supply-demand ecosystem ↗

Han Xinyi believes that the supply and demand sides of agents have not yet formed a positive cycle. Ant Group will play the role of a "catalyst," breaking the deadlock by solving trust issues and driving the ecosystem flywheel. To connect user needs with merchant services, Alipay's "Abao" will introduce incentive policies to encourage developers to create higher-quality agents.

Nvidia and Palantir reach cooperation to bring sovereign AI into critical supply chains ↗

Palantir and Nvidia Announce Collaboration to Bring Sovereign AI Technology to Supply Chain Management, First Implemented in Nvidia's Own Factories. The Solution Combines Nemotron Open-Source Models with the Palantir Foundry Platform, Aimed at Achieving Machine-Speed Decision Support and Ensuring Stable Supply of 1.3 Million Components.

Not only Anthropic's Mythos 5, EU institutions have been granted access to OpenAI GPT-6 Astra model ↗

The European Union Agency for Cybersecurity (ENISA) has been approved to test Anthropic's Mythos 5 and OpenAI's GPT-6 Astra models. Previously the White House restricted foreign institutions from accessing high-performance AI; this EU approval may be an exception.

After months of negotiations, Anthropic opens Mythos 5 access to EU ENISA ↗

After months of negotiation, EU cybersecurity agency ENISA finally gained access to Anthropic's high-performance AI model Mythos, but cannot use the latest version. Previously, the White House had restricted foreign institutions' access. This opening aims to strengthen cybersecurity vulnerability screening.

Alibaba Plans to Lead a $300 Million Funding Round for AI Evaluation Company UniPat, Founded by a Former Intern, Valued at $2.5 Billion ↗

UniPat was founded by Li Kuan, who last year worked at the Tongyi AI Lab on post-training analysis, data synthesis and reinforcement learning, and designs testing and evaluation scenarios close to real usage environments for AI models. The company specializes in generating detailed training and evaluation data, and plans to sell to other researchers and enterprises in the AI field.

Alipay establishes"Agent Emergence Award": An annual investment of 10 million; no coding skills required to participate ↗

Alipay announced the establishment of the 'Agent Emergence Award', investing 10 million development fund each year, with an annual grand prize of up to 1 million. Those who cannot code can also participate, generating applications through conversation. Award-winning applications also have the opportunity to access Abao's service for 1 billion users. #AlipayAgentEmergenceAward# First-phase shortlist gets 5000 yuan reward.