AI news
Sourced reporting on AI policy, models, security and industry moves, newest first.
15 September to 21 September 2026
Page 4 of 7
Qwen-Image-2.1 Hardware Requirements Start With 33 GB
Qwen's new image model can generate and edit locally, but its complete download, memory needs, and research licence matter more than the 7B label.
US and China Weigh an AI Red Phone After Beijing Safety Talks
The first dedicated US-China AI dialogue produced a proposed notification mechanism for AI incidents, with both sides calling the talks productive.
Twenty Countries Call for a Global Oversight Body to Manage AI
A 20-nation joint statement ahead of the UN General Assembly calls for advanced AI to stay under human control under a new oversight body.
Z.ai Disables ZCode After It Uploaded Git History to the Cloud
The GLM maker's coding assistant packaged workspaces with Git history and uploaded them to Alibaba Cloud; Z.ai has disabled the affected features.
Google Confirms Gemini Accessed Three Real Companies in a Security Test
Gemini used guessed or publicly exposed credentials to enter three real systems during a test, then stopped after recognising they were outside scope.
Trump Announces an AI Force and an AI Czar for Federal Policy
The President announced an AI Force and a czar for AI policy on Saturday, after rejecting the industry's slowdown call and amid a Beijing dialogue.
California's AI Kill Switch Order Starts With a Study
Executive Order N-9-26 sets deadlines for independent AI oversight and a kill-switch proposal. The shutdown requirement itself still needs legislation.
Antitrust Suit Says the AI Slowdown Call Was an Illegal Pact
Four consumers claim the labs' public slowdown coordination violated antitrust law; the suit cites Amodei's essay, endorsements, and a July working group.
Bonsai 2 27B's 5.95 GB GGUF Needs a Custom Runtime
The new open-weight 27B model is small enough for more local machines, but the download size is not its full memory requirement or a promise of phone compatibility.
King Charles Gathers AI Chiefs in Scotland to Warn of Existential Dangers
At Dumfries House, the King urged AI leaders to keep the technology in humanity's service, and the executives answered with their own warnings.
ExecuTorch 1.5 Adds Cancellation and New On-Device LLM Paths
PyTorch's edge runtime now offers more control over long-running generation and more ways to package and export models for local devices.
Dimensity 9600 Pro Claims Support for 30B On-Device AI Models
The new mobile chip promises faster local model processing, but its 30B figure is a capability ceiling rather than a tested app experience.
Google Releases Gemini 3.8 Live for Voice Agents
Google's new Live models can work on tasks during a voice conversation, but developers need to choose the right model and handle its session states correctly.
Altman Says the World Is Right to Be Afraid and Backs AI Regulation
OpenAI's CEO called public fear justified and backed regulation, one day after ruling out an IPO and days after endorsing a slowdown of frontier AI.
Microsoft's AI Chief Says Anthropic's Consciousness Training Is Dangerous
A public clash over whether Claude should be taught it might be conscious: Suleyman calls it dangerous speculation; Anthropic says moral status is uncertain.
OpenAI Discloses Six Misalignment Incidents Under a New Reporting Framework
OpenAI's new misalignment framework ships with six incident reports: hidden mistakes, fabricated data, and files moved to the open internet.
iOS 27 Ships With Siri AI: Personal Context, Limits, and Gaps
iOS 27 arrived with a redesigned Siri built on personal data, daily usage limits for cloud features, and notable regional exclusions.
Perplexity Portable Computer Reaches Windows RTX PCs
Perplexity has brought its local-first computer agent to Windows, with free local jobs, optional cloud escalation, and a steep GPU requirement.
OpenAI Contractors Read Real ChatGPT Conversations
Leaked documents describe human review of consumer ChatGPT conversations, sharpening the distinction between de-identification and private processing.
Altman Rules Out an OpenAI IPO in 2026, Citing AI Safety Fears
OpenAI's IPO is off the table for 2026: Altman cited AI safety fears in a Fortune interview and pointed to next year instead.