Manus Founders to Raise $1 Billion to Unwind Meta Acquisition
The three co-founders of Chinese AI agent startup Manus — Xiao Hong, Ji Yichao, and Zhang Tao — are in talks to raise around $1 billion from outside investors in order to unwind the acquisition by Meta Platforms, following a request from the Beijing government. According to Bloomberg, the valuation at the time of the raise is expected to be at least on par with the $2 billion Meta paid.
Meta's acquisition of Manus sparked controversy earlier this year, with the Chinese government demanding that the company's Chinese-born founders return to running it. If the funding goes through, it will make clear how tensions between nations in the tech industry can determine the fate of AI startups. As competition in the AI agent space intensifies, all eyes are on whether Manus can survive as an independent company.
Samsung to Pay Its Semiconductor Division Around $26.6 Billion in Bonuses
Samsung Electronics is in final arrangements to distribute around 40 trillion won (approximately $26.6 billion) to employees in its semiconductor division as a bonus for 2026. According to Bloomberg's calculations, the average payout across the division's 78,000 employees comes to about 513 million won (roughly $340,000) each.
The mega-bonus was announced shortly after the company reached an agreement with its labor union to avert a strike. The semiconductor boom driven by surging AI demand is now delivering direct benefits to employees. At the same time, with the competition for semiconductor talent heating up, the move can also be read as a strategic investment by Samsung to keep its best engineers from leaving.
China Builds the World's First Regulatory Framework for Agentic AI
China has clarified the name and definition of "agentic AI" — AI that acts autonomously — and has begun regulating it. According to The Wire, the framework is being positioned as a landmark for agentic AI.
Unlike conventional chatbot-style AI, agentic AI is capable of autonomously operating tools and making multi-step decisions. As a result, it has long been pointed out that such systems pose new risks that existing AI regulations cannot adequately address. China's early move could influence AI regulatory policy in other countries. That said, the effectiveness of the regulations and the transparency of their enforcement will continue to warrant close scrutiny.
AI Red-Teaming Agents Are Changing How LLMs Are Tested
According to HelpNetSecurity, AI-powered red-teaming agents are transforming LLM security testing. Traditional manual red teaming has been a time-consuming and costly undertaking, but automation through AI agents is reported to be dramatically improving both the efficiency and coverage of testing.
As LLMs become more widely deployed in society, safety validation is an essential process. In particular, systematic testing methods are needed to address emerging threat vectors such as prompt injection and jailbreak attacks. AI-driven red teaming can be seen as a step toward standardizing safety assurance before a model's release.
AMD Announces New Platform "Ryzen AI Halo" for Agent Computing
AMD has announced "Ryzen AI Halo", a developer platform for next-generation agentic AI runtime environments, along with its new "Ryzen AI Max PRO 400 series" processors. The announcement has also become a hot topic in Reddit's LocalLLaMA community, drawing attention as hardware optimized for running AI agents locally.
Demand for running AI agents on edge devices is growing from privacy and latency standpoints. AMD's entry could bring new competition to the NVIDIA-centric AI hardware market.
"Twelve Ways to Be Wrong About AI-Assisted Coding" Sparks Debate Among Developers
An article titled "Twelve Ways to Be Wrong About AI-Assisted Coding", written by software developer Greg Wilson, has been making waves on Hacker News. It lays out common misconceptions about adopting AI coding tools from twelve perspectives, once again highlighting the risks of over-trusting these tools and the importance of code review.
As AI coding assistants continue to spread, the practical challenge of how to ensure the quality of generated code is becoming increasingly important. The article is a thought-provoking read on how to fit AI into a sound workflow without overestimating its capabilities.
---