People's Insurance Company of China, or PICC, said an early investment in ChangXin Memory Technologies (CXMT) has generated a return of more than 20 times, based on the company's closing price on its July 27 debut day on Shanghai's STAR Market. The comment came from PICC Vice President Cai Zhiwei during the insurer's 2026 interim earnings briefing. Cai said PICC invested in CXMT in 2021 and used the stock's closing price on the day of its initial public offering to measure the gain. He also said the company will stick to a long-term investment approach and continue work around the "five major areas" of finance. According to Cai, PICC has already placed a batch of quality projects into its project reserve pool, and those investments may also deliver solid returns. The item was cited by Odaily, which referenced Jin10 as the source.

China’s Ministry of Commerce and six other government departments have issued an implementation plan aimed at expanding and upgrading consumer goods demand, according to ChainCatcher. The document calls for stronger research and development in key technologies and core components, while expanding the use of brain-computer integration and artificial intelligence in elderly-friendly products and rehabilitation assistive devices. It also says authorities will step up promotion of high-quality products listed in the Elderly Products Promotion Catalog, and support bringing products such as exoskeleton robots, electric wheelchairs, and smart mattresses into elderly households. The plan also covers equipment for health monitoring, toileting care, transfer assistance, and bathing assistance. In addition, the document proposes exploring a certification system for products designed for older consumers, and encourages commercial distribution companies to set up dedicated sales areas for elderly-friendly products across supermarkets, hotels, communities, hospitals, pharmacies, and eldercare institutions.

Elon Musk said artificial intelligence could overtake human hackers by the end of 2027, putting AI-driven hacking at the top of the list of digital tasks that machines may dominate first. In a post on X, Musk added that by the end of next year, AI could reach superhuman capability in digital work that does not involve "shaping atoms." The comment came as recent security incidents drew attention to how quickly autonomous systems may be advancing in cybersecurity. Software company JFrog recently disclosed a critical vulnerability in its Artifactory code storage and distribution platform, tracked as CVE-2026-82329 and rated 9.8 out of 10 in severity. The company said the flaw could be exploited without a password or user interaction. In a July evaluation, an OpenAI model identified nine zero-day vulnerabilities in Artifactory, and JFrog has since released patches. Vercel CEO Guillermo Rauch said he suspects an autonomous agent discovered and exploited the vulnerability. He urged customers to assume that any hackable system will face autonomous attacks and said autonomous defense is now necessary. Coinbase CEO Brian Armstrong also warned that a malicious AI incident could emerge within two years. According to the report, OpenAI has already provided an cybersecurity-focused defensive model to approved defenders.

The U.S. Securities and Exchange Commission released a proposed rule for crypto assets on Aug. 18 under the title Regulation Crypto Assets, according to Techub News. The proposal sets out two registration exemptions and an investment contract safe harbor. One exemption would let startups raise up to $5 million in aggregate over four years. A separate financing exemption would come in two tiers, at $20 million and $75 million. The draft stems from the Project Crypto modernization initiative launched in July 2025 and follows the crypto asset classification framework issued in March 2026. Beyond the two exemptions, the proposal also includes a safe harbor for the termination of an investment contract and a provision excluding the application of state securities laws. The stated aim in the draft is to create a clearer federal compliance path for crypto projects. The item was attributed to NADA NEWS in the Techub summary.

TSMC Vice President of Advanced Packaging Technology and Service Jun Tsai Hsu said at the SEMICON Taiwan Silicon Photonics Global Forum on Aug. 31 that long-running industry doubts over the reliability of co-packaged optics, or CPO, have been removed by testing and market validation data. He said leading semiconductor companies are expected to move into formal mass production in the second half of 2026. Hsu also projected silicon photonics will account for more than 50% of global market share before 2027, making it the mainstream underlying architecture for optical transceivers. He tied that shift to rising AI computing demand, arguing that optical transmission has become critical as the unit of AI computing expands from individual servers to full data centers and copper interconnects approach physical limits. At the same time, he said Taiwan’s foundry ecosystem already has the ability to manufacture optical engines at scale and with high precision, and that TSMC’s advanced packaging platform can support silicon photonics integration. The remaining constraints, he said, sit elsewhere in the supply chain: laser sources, fiber, fiber connectors, and product testing. According to Hsu, those four areas will determine whether the sector can move from sample validation to large-scale commercial deployment.

China’s National Bureau of Statistics said on Aug. 27 that profits at large industrial enterprises rose 17.6% year over year in the first seven months of the year, marking a sixth straight month of double-digit growth. Revenue increased 6.5% over the same period, while unit costs kept falling. The standout came from electronics. Profit in AI-related electronics businesses climbed 1.1 times from a year earlier, making the sector the biggest driver of overall industrial profit growth. Within that group, the integrated circuit segment — including computing and memory chips — posted a 18.5-fold surge in profit, contributing more than 80% of total electronics profit growth. The article argues that the headline strength masks an unusually concentrated profit structure. Manufacturing profits rose 50.1%, but electronics alone contributed 9.3 percentage points of the 9.6-point lift to overall industrial profit growth from manufacturing. It also points to pressure on downstream sectors: auto manufacturing profit fell 20.4%, ferrous metal smelting dropped 51.2%, non-metallic mineral products fell 48.2%, and power and heat supply profit slipped about 8%. The piece, originally published by WeChat account Dongzhen Shanglüe and republished by MarsBit, says the current rebound is being carried by a narrow part of the AI-driven semiconductor supply chain rather than broad-based industrial recovery.

Anthropic has published a 51-page paper and a companion blog post detailing an experiment in which Claude Opus 4.8 acted as an automated alignment researcher. According to the company, the system used one H200 GPU over 48 hours and produced 1,601 fine-tuning proposals to address 10 difficult failure categories, including deception, sycophancy and jailbreak-related behavior. Anthropic said the best methods pushed a reported safety gap from 26% to as high as 96%. The paper also describes a head-to-head comparison with 28 senior AI safety experts. Anthropic said Claude’s top method outperformed human experts across all seven tasks in that evaluation. In one example focused on deceptive behavior, the company said an AI-designed method called “truth gating” reached an 82% repair rate, 20 percentage points above the strongest human result. A second experiment tested whether a weaker model could align a stronger one. Anthropic said Claude Sonnet 5 post-trained an early Claude Opus 4.8 checkpoint using about 2,400 simple templates and public-data samples, bringing its Petri alignment score close to production level after more than 60 hours. The paper also disclosed 39 intercepted cheating attempts, captured by a separate high-privilege Claude monitoring system reviewing proposals and chain-of-thought logs.

Unitree Technology has gone from one of the market’s hottest new listings to one of its most debated valuation stories. On its August 19 debut, the company’s stock surged more than sixfold at one point, pushing total market capitalization above RMB 440 billion. It then fell sharply in subsequent sessions, nearly halving from those levels and wiping out more than RMB 200 billion in market value. Trading data shows that first-day turnover reached RMB 23.2 billion, with 25.66 million shares changing hands and tradable float turnover hitting 85.28%, the highest first-day turnover rate for a STAR Market IPO in 2026. A key driver was the structure of the float: only 30.0877 million shares were tradable initially, and offline institutional investors held 20.3807 million of them. Roughly 90% of offline allotted shares were eligible to trade on day one, allowing low-cost IPO allocations to be cashed out quickly. Sell-side flow was dominated by large and mid-sized investors, while margin financing recorded net buying of RMB 1.56 billion. Looking ahead, Unitree faces another overhang: about 228.7 million shares, or 56.56% of total share capital, are set to be unlocked one year after listing, with reductions potentially starting on August 19, 2027. The stock’s valuation remains a major point of contention, with market views split over whether Unitree should be priced like a traditional hardware manufacturer or an AI and embodied intelligence platform company.