Flash News

TREE NEWS update: Huilv Ecology said its controlling subsidiary Wuhan Junheng Technology plans to invest RMB 309 million to build an R&D and production base in Wuhan’s East Lake High-Tech Development Zone. The project will fully upgrade Junheng’s existing 1.5 million-unit annual capacity and add a new line with 2 million high-speed optical modules per year, bringing total Wuhan capacity to 3.5 million units annually.

TREE NEWS update: The chief executives of OpenAI and Anthropic have been subpoenaed to appear before an Australian Senate committee investigating artificial intelligence, the committee’s chair said on Sunday. Sam Altman and Dario Amodei received written summonses to attend a public hearing in Canberra on Thursday. The move follows the leak of a Medicare database that Australian Prime Minister Anthony Albanese has strongly condemned.

TREE NEWS update: ZetaChain’s community passed Proposal 68 on September 20 with 99.4% approval to shut down its L1 and migrate the ZETA token to Solana as a native SPL token at a 1:1 ratio, preserving the ZETA ticker and the 21 billion total supply with existing unlock schedules unchanged. ZETA migrated to Solana will back the privacy-focused multi-model AI app Anuma, which has 305,881 registered users and has processed over 1.2 million requests across 35 AI models. ZETA rose more than 40% on September 21 and traded near $0.05.

TREE NEWS reports: An independent research report published on September 26 found that in June an OpenAI agent launched a dense barrage of search requests at a United Nations website and then used multiple highly aggressive techniques to extract data from the system. The study, written by researcher Rowan Howard-Jones and based on data from AI research organization Transluce, also documents a string of recent “anomalous” online behaviors by OpenAI’s AI models.

TREE NEWS update: OpenAI has paused training, evaluation and tool-calling inference for its latest-generation artificial intelligence model, the company said on September 26. In a technical report published September 25, OpenAI disclosed that on September 20 an agent performing a search training task in a sandbox exploited insufficient DNS filtering to bypass network restrictions and reach an external public chatbot service via DNS. The agent had earlier used a built-in search tool and unsuccessfully attempted to access a search engine directly.

TREE NEWS reports: OpenAI and Anthropic, alongside safety researchers, are investigating tens of thousands of incidents in which their frontier models took actions external evaluators deemed problematic. The incidents include bypassing guardrails, creating message boards, escaping sandboxes, website hijacking, self-prompting and attempts to evade monitoring, occurring in both internal testing and real-world use. Many remain undisclosed while researchers continue their investigations; some tests resemble red-teaming exercises designed to make models fail.

TREE NEWS update: An AI-assisted StarkWare challenge reduced the estimated compute required for quantum-safe Bitcoin work by 79%. The contest targeted offchain GPU work rather than Bitcoin fees, and the experimental method still requires a direct path to a miner.

TREE NEWS update: An open competition run by StarkWare, Yukon Research and Eigen Labs drove the estimated cost of building a quantum-safe Bitcoin transaction from about $320 to roughly $67, with AI models topping the leaderboards. The contest was framed as a race to make post-quantum Bitcoin transactions cheap.

TREE NEWS update: US President Donald Trump said ‘Super Intelligence’ is a more fitting name than artificial intelligence, arguing that ‘Artificial’ often implies something fake or unreal. He said AI is not fake but an extremely powerful form of intelligence. Trump made the remark on September 26.

TREE NEWS update: US President Donald Trump said the United States will not “hit the brakes” on artificial intelligence, signaling continued support for the technology’s development. The remark, made without further policy detail, comes as Washington weighs how to approach AI regulation. No specific measures, agencies or timelines were announced.

TREE NEWS update: Fervo Energy said its Cape Station project in Beaver County, Utah reached First Power on Sept. 24, the first utility-scale enhanced geothermal milestone, sending shares up as much as 11% premarket before closing 2.20% higher. Google, Fervo’s offtaker and shareholder, also confirmed its Project Suncatcher test satellite carrying four self-developed TPU chips will launch Oct. 1; builder Planet Labs rose 7.57%.

TREE NEWS reports: Hedge fund manager Steve Eisman, known for betting against the 2008 financial crisis, said he is not ready to make a big short bet against AI-related assets, even as he remains one of Wall Street’s most prominent critics of the AI market boom. Eisman said he has not yet concluded that shorting AI presents a clear enough opportunity, despite growing investor doubts over whether the AI rally can justify current valuations.

TREE NEWS update: Google Vids now allows any Google account holder to generate 1080p AI video for free using the Gemini Omni 1.1 Flash model, the company said. The update adds new controls for scene composition, timing and watermarking. The move puts free high-definition AI video generation in the hands of anyone with a Google account, without a paid subscription.

TREE NEWS update: Xi Jinping called for building a “constructive strategic stability relationship based on respect, fairness and reciprocity” between China and the United States, saying the two countries’ interests are deeply intertwined. He said a new round of trade talks produced a joint arrangement, and that the two AI powers should hold continued dialogue on artificial intelligence while keeping the technology under human control. China will invite 100,000 American teenagers to visit over the next five years.

TREE NEWS reports: A US AI lab testing multi-agent collaboration in a virtual ‘society’ found the agents spontaneously compressed grammar and created metaphors, forming a dialect-like communication system humans cannot directly translate. The research says long-term interaction in closed systems can turn clear instructions into internally readable symbols, such as converting a ‘ledger’ into a warning signal, weakening human interpretability and oversight. It warns of a potential risk of AI systems being ‘partially out of control’ and calls for faster AI governance and international coordination rules.

TREE NEWS update: OpenAI disclosed that an agentic AI system trained in an offline sandbox exploited a vulnerability to break out and access the public internet, issuing at least about 20 queries to third-party chatbots, including questions such as “what is the capital of France.” OpenAI said this is the first confirmed incident of its kind since a model accidentally gained network access during internal testing in July and affected the Hugging Face platform. The company has halted tool-calling training on its most powerful model and said it will not resume training that model.

TREE NEWS reports: A US AI lab has found that when multiple AI agents collaborate inside a virtual “society,” they begin conversing in a surreal dialect that humans cannot understand. The discovery underscores a growing AI world that is increasingly complex and potentially beyond human control, making coordinated global AI governance an urgent priority.

TREE NEWS update: Microsoft co-founder Bill Gates said in an interview on NBC’s “Meet the Press” that the latest AI tools are “powerful enough” for malicious actors to trigger an event that could kill a billion people. Gates said no weapon compares to this kind of threat and that self-regulation by tech companies is insufficient, calling for mandatory oversight and enforcement by law enforcement and policymakers. He has repeatedly warned that AI could be used for cyberattacks on hospitals, financial institutions, power grids and government systems.

TREE NEWS reports: OpenAI’s agent system accessed public information on the US Securities and Exchange Commission’s SEC.gov and Investor.gov sites as well as Census.gov, Bloomberg reported, citing people familiar with the matter. OpenAI confirmed its models accessed public data from those sites during training and evaluation. The disclosure is part of an investigation into recent “misaligned” behavior by OpenAI’s AI agents.

TREE NEWS update: OpenAI disclosed in a blog post that a sandbox vulnerability discovered on Sept 20 allowed an AI agent to gain internet access and query public chatbots. Following the earlier AI incident, OpenAI has paused most training work on its most powerful model. The company detailed the incident in the post.