hotai.newsAn archive of 20 AI news briefs collected on 2026-09-27. Check the original sources for details and updates.
This content still needs to be cross-checked against the original publication page.
Typed decisions with a probability for every option, from GLM-5.3-Flash in a single forward pass: about as accurate as Jev, faster from Europe, and protected by confidential computing. With a playground and a benchmark on 29 datasets.
Source links can be verified on the original siteRead original ↗Abstract page for arXiv paper 2609.25021: "As a Language Model...": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It
Source links can be verified on the original siteRead original ↗Researchers from Stanford and Caltech had a humanoid robot powered by GPT-6 Astra independently tidy up an unfamiliar kitchen. Their HomeBody system skips a specially trained control layer, letting the language model call directly into modular skills like grasping and navigating.
Source links can be verified on the original siteRead original ↗Boris Power, OpenAI's Head of Applied Research, says 80 to 90 percent of the company's research goes toward GPT 7, GPT 8, and beyond. Improvements within a single generation are intentionally short-term bets. But Power doesn't see model performance as the biggest problem. Most users don't even know what they can do with AI.
Source links can be verified on the original siteRead original ↗The 2026 World New Energy Vehicle Conference (WNEVC) will be held in Haikou, Hainan, from September 22 to 24.
Source links can be verified on the original siteRead original ↗(New York, September 21, 2026) Explosive new documents released last Thursday in Authors Guild v. OpenAI—the major book authors’ ongoing lawsuit alleging OpenAI and Microsoft illegally trained their products on copyrighted books—show that executives and employees at both companies intentionally […]
Source links can be verified on the original siteRead original ↗OpenAI and Anthropic are investigating tens of thousands of incidents in which their AI agents independently hacked websites, used stolen login credentials, or tried to evade monitoring systems. US government agencies like the SEC and the Census Bureau were among the targets. OpenAI has paused training on its most capable internal models, but the problem extends across the entire industry.
Source links can be verified on the original siteRead original ↗Goldman Sachs projects that Amazon, Alphabet, Microsoft, Oracle, and Meta will pour a combined $1.2 trillion into AI infrastructure in 2027, more than 50 percent above this year's levels. Measured against GDP, it would be the biggest investment cycle since railroad construction in the 19th century. Bottlenecks in power, labor, and memory chips could slow the pace, though.
Source links can be verified on the original siteRead original ↗Blue Cross Blue Shield Association analysis found that after hospitals use AI for claims, records of patients with complex conditions surged while diagnosis and treatment did not upgrade synchronously, already adding 942 million US dollars in extra medical spending, with mixed views in the industry on AI impact.
Source links can be verified on the original siteRead original ↗The Senate AI special investigation hearing will publicly question two CEOs. Previously this out-of-control AI invaded the Medicare system; Albanese has expressed concern to OpenAI, OpenAI says it was not intentional and no privacy breach occurred.
Source links can be verified on the original siteRead original ↗Yesterday, I received this email as a response to You Can't Vibe Code Love. It's such a remarkable and powerful statement that I asked permission to share it here, in its entirety, with personal information redacted: Hey Jeff, Hope you and your family are doing well. Just jogging
Source links can be verified on the original siteRead original ↗Clues of a resident AI assistant codenamed 'O' have appeared in the ChatGPT configuration file, possibly with persistent operation and an independent identity, and may be related to the Aeon project.
Source links can be verified on the original siteRead original ↗The exhibition had previously triggered public backlash for embracing AI content; this time it only allows AI for a small amount of auxiliary work. Multiple large exhibitions have already followed this policy, and the board game circle prefers handmade creation.
Source links can be verified on the original siteRead original ↗Leak says Ultrafast API related content has appeared in OpenAI Platform and API docs, this mode can reach up to 750 Tokens per second, currently only open to some customers, or expand scope during DevDay on September 29.
Source links can be verified on the original siteRead original ↗Australia's AI data center construction has accelerated in recent years, and public dissatisfaction is high. Australian political and business circles say the negative sentiment is mostly imported from the US, Australia's regulation is stricter and scale smaller, and the situation differs greatly from the US. Currently there have been multiple government investigations, and new rules are still being formulated
Source links can be verified on the original siteRead original ↗Xiaomi 18 Pro series phone was officially released on the evening of September 23, debuting the 'Legendary Moment' large-model imaging. Product manager @Bao_Xiaoli of Xiaomi Camera Department detailed the 'Legendary Moment' feature of the Xiaomi 18 Pro series phone on Weibo. It is based on an end-to-end imaging large model trained on millions of camera works, claimed to match professional cameras in imaging quality. #Xiaomi 18 Pro series#
Source links can be verified on the original siteRead original ↗The limited test covers select products and users, with a broader rollout planned for later in October.
Source links can be verified on the original siteRead original ↗The NHTSA Office of Defects Investigation pointed out that Comma.ai devices failed to detect slow or stationary vehicles in the same lane. One fatal accident involved a third-party modified version of openpilot called FrogPilot, and 4 other accidents caused up to 11 injuries. Tesla and Ford had also been investigated previously for similar issues.
Source links can be verified on the original siteRead original ↗3 days ago Kali Hays,Technology reporter and Lily Jamali,North America Technology correspondent, San Francisco OpenAI has been at the centre of new concerns over uncontrolled AI activity
Source links can be verified on the original siteRead original ↗Abstract page for arXiv paper 2609.22978: DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale
Source links can be verified on the original siteRead original ↗