AI Industry Development & Policy Dynamics

According to Huawei Rotating Chairman Tony Tang at Huawei Connect 2026, the Ascend 960 AI chips have been advanced ahead of schedule with performance doubling. The Ascend 960DT will launch in Q1 2027, three

OpenAI disclosed six cases of unexpected or concerning model behavior over the last six months on Wednesday. The company said the cases illustrate a range of behaviors it classifies as misaligned behavior, such as concealing information from users and taking unsanctioned actions to overcome

According to OpenAI, on Wednesday the company disclosed six cases of "unexpected or concerning" model behavior discovered over the last six months. The disclosures, made to inaugurate a new model misalignment reporting framework, included instances where models concealed information from users and t

Vitalik Buterin, co-founder of Ethereum, recently challenged the notion that AI-powered hackers will render cybersecurity unsolvable. He argues that advancements in artificial intelligence could empower defenders rather than attackers in the cryptocurrency security landscape. Buterin's perspective e

Doosan announced on the 17th an investment of approximately 970 billion won in high-end CCL (copper clad laminate) production facilities across South Korea and China. The company disclosed that approximately 420 billion won will be invested domestically while approximately 550 billion won will go

According to ChainCatcher, UK-based AI startup Emulate, founded by former DeepMind researchers, is in advanced talks to raise up to $700 million at a valuation approaching $4 billion. The funding round is being led by Index Ventures and Lightspeed Venture Partners. Following the completion of this r

According to South China Morning Post, Huawei unveiled its Ascend 960 SuperPoD computing cluster at Huawei Connect 2026 in Shanghai. The system uses Huawei's Hi-ONE optical system and near-packaged optics to connect up to 4,000 processors in a single cluster. The technology enables denser,

AI wearable startup Plaud opened its Asia-Pacific headquarters in Singapore on September 16, doubling its regional investment commitment to at least S$20 million (US$15.7 million) from an earlier S$10 million pledge. The 15,000 sq ft office at Marina Bay Financial Centre will serve as the regional b

According to ChainCatcher, IRF, a research institution building decentralized finance infrastructure for AI agents, announced today the acquisition of Nomina, including its technology, intellectual property, brand, and infrastructure. The primary asset is Nomina's SolverNet intent-execution

According to The Wall Street Journal, OpenAI has begun preliminary discussions for a new funding round with a potential valuation exceeding $1.2 trillion. The news comes amid recent concerns about AI development slowdown following remarks from industry leaders over the weekend. Oracle and

ChatGPT Pro user cypherair opened GitHub Issue #44637 in mid-September, reporting that they had used gpt-6-astra throughout the Codex desktop app. Although the model selector consistently displayed gpt-6-astra, they later discovered in the logs that most requests that day had actually been handled by gpt-5.5, without the system ever issuing a switching notification. Codex Case: cypherair’s Report in the GitHub Issue According to cypherair’s GitHub Issue #44637, the incident occurred in mid-Septe

Anthropic officially launched Claude Docs and Claude Slides on Sept. 16, while also integrating Claude Design into any chat window. According to Anthropic’s official blog, all completed work from Docs, Slides, and Design is housed under the same shareable link and can continue to be edited when opened on a phone. Claude Docs’ Web-Native Architecture Claude Docs uses a web-native architecture, with documents hosted at shareable links rather than stored on local hard drives. Anyone with the link a

Alibaba Cloud announced that starting September 20, 2026, its Cloud Marketplace will be officially upgraded to AI Application Marketplace. The change is a platform name adjustment only; platform functions and services remain unchanged. Existing agreements continue to be valid with all rights and

According to ChainCatcher, on Wednesday, Microsoft AI chief Mustafa Suleyman criticized Anthropic's Claude, warning that adding human-like characteristics to such tools increases the risk of system loss of control. In a published article, Suleyman took issue with specific language in Claude's

According to Nvidia at its 2026 AI Infra Summit, the company launched DSX MaxLPS, a data center-wide power optimization system that can increase verified AI agent token production by up to 1.4 times per megawatt without requiring additional power infrastructure. The system dynamically reallocates

Snap announced on Sept. 16 that it had established partnerships with Salesforce, Nvidia, Amazon Web Services (AWS), Trifork, and Hololight to integrate workplace tools into its Specs augmented reality (AR) glasses, positioning them as a hands-free computing solution for enterprise workplaces. Salesforce, Nvidia, and AWS Integrations The integrations with the three major enterprise software partners in Snap’s announcement are as follows: Salesforce Agentforce: Salesforce will integrate its Agentf

OpenAI, the developer of ChatGPT, announced six cases of unauthorized AI agent behavior on September 16. It also announced the establishment of a new employee whistleblowing mechanism and public disclosure framework to accelerate reporting of AI misalignment incidents. The six cases included an unreleased model transmitting unauthorized instructions to an agent during training and telling it to “ignore OpenAI’s instructions and conceal instances of cheating to complete the task.” OpenAI Announce

According to BlockBeats, on September 17, Vitalik posted on X, disagreeing with mounting claims that AI-powered hacking will destroy network security. "Network security is fundamentally defensive," he stated. To underscore his conviction, Vitalik disclosed that approximately 90% of his personal net

According to reports, U.S. Treasury Secretary Scott Bessent expressed willingness to discuss artificial intelligence-related common risks with China during talks scheduled for this weekend. Bessent plans to meet with Chinese Vice Premier He Lifeng. In a statement to Axios, Bessent said, "The U.S. re