Twins Lee and Lucas Brown, founders of Tune, launched the Alaska News platform, where AI agent Walter generates local news from public documents. Since the start of the year the agent has written more than 4,500 items using Anthropic, OpenAI, Google and xAI models, under human editing.
SAP is integrating n8n visual workflow orchestration into Joule Studio, an environment for building multi-agent scenarios on SAP data and external services. General availability is expected in Q3 2026; pricing and terms are undisclosed.
The open-weight Kimi K3 model (2.8 trillion parameters) became available on Amazon Bedrock on September 18, 2026. It supports 1 million token context, native vision, and offers ~2.5x scaling efficiency gain over Kimi K2. According to the developer, the model trails Claude Fable 5 and GPT 5.6 Sol.
SAG-AFTRA and WGAE said AI companies' warnings about existential threats distract from real issues: replacing actors with digital copies and job losses. The unions noted they secured contractual guarantees and compensation for AI doubles after the 2023 strikes.
Canadian firms Cohere and OpenText are teaming up to offer an agentic AI platform to OpenText's approximately 120,000 enterprise customers. The North platform and Cohere models will connect to OpenText data; the solution will launch early next year.
A Politico poll of more than 2,000 adults found about two-thirds of Americans consider the chance of advanced AI destroying humanity at least moderate. 17% are almost certain of that outcome, and 48% support a full pause on AI development.
Nepal authorities are using the FaceTagr facial recognition system alongside DNA analysis to identify flood victims. Of 1,377 bodies found, only 102 (7%) were identified by September 11, while two DNA centers process fewer than 1,000 samples a year.
Tether AI Research introduced the open TranslatePsy-EuroNano model family for on-device translation between English and nine European languages. Deployment takes 36–89MB versus 633MB for 18 bilingual Firefox Bergamot models, with comparable quality.
Tether introduced TranslatePsy-AfriSLM, open machine translation models covering 19 Sub-Saharan African languages. An 800M-parameter model outperformed Qwen3.5-122B-A10B, TranslateGemma-27B and NLLB-3.3B on three benchmarks and runs on-device without cloud.
An open-source benchmark was released to evaluate AI agents' ability to design working robots, not just write code. The test checks how agents handle tasks in the physical world.
OpenAI, Anthropic, and Google DeepMind have held closed-door talks for weeks on unified AI safety standards and granting independent evaluators access to the development process. Organizations like METR and Apollo Research are planned to get access to intermediate model checkpoints to detect deceptive behavior. Sam Altman and Demis Hassabis backed the initiative after Dario Amodei's essay.
DoorDash built a multi-agent LLM system to automatically remove stale feature flags. In a test on 50 flags, the system produced working pull requests for 45 of them, spending an average of 13.8 minutes and $4.79 per cleanup versus 1–2 hours of manual work.
AWS introduced Amazon SageMaker HyperPod Inference Gateway, a Kubernetes-native router for LLM inference that is aware of GPU state. According to AWS, time to first token drops by up to 82%, and P95/P99 latencies fall by 94–98% under mixed workloads.
A developer under the nickname Uzbekunknown posted the dlss-nr-on-intel project on GitHub, a port of Nvidia's DLSS 5 neural rendering to Intel Xe architecture. The 71-block U-Net model was rewritten for Arc 140V XMX blocks via the Vulkan VK_KHR_cooperative_matrix extension and runs in FP16. In Tekken 7 tests at 640x360 it delivers about 10.5 FPS, while a full DLSS 5 pass in Full HD takes 412 ms.
Growing alarm over AI risks has brought little-known safety research groups into the spotlight. Dario Amodei (Anthropic) and Sam Altman (OpenAI) promised to embed external evaluators in their companies; Satya Nadella (Microsoft) also supported the idea. Critics doubt the independence of such groups.
More than 100 AI specialists, including Geoffrey Hinton, signed an open letter demanding that independent safety evaluators get access to models from leading labs. They insist on scientific objectivity, transparency, and protection from corporate pressure.
Gartner described four levels of AI in warehouse automation: advanced optimization models, generative systems, semi-autonomous agents, and physical robotics. Logistics has moved from pilots to real deployment due to staff shortages, lower startup costs, and algorithm maturity.
US AI developers and authorities are concerned about 'distillation attacks'—training models on outputs from other frontier models—allowing China and Russia to create similar systems more cheaply. China rejected the accusations and promised 'countermeasures' if the US uses this as a pretext to constrain its AI. The topic may be raised at the US-China leaders' meeting on September 24.
A joint German-UK study found AI models trained on anonymized medical records can memorize patients and carry their past conditions into current predictions. In one case, a model gave only a 15% heart attack probability versus 73% for a model without historical data.
OpenAI is ending support for Custom GPTs, custom ChatGPT versions with their own instructions and files. For Enterprise workspaces, creating new GPTs will be blocked on September 25, and existing ones will stop working on December 11; other tiers may follow the same schedule.
A Global Undergraduate Awards study found that 55% of students consider AI a significant concern, up from 48% in 2025. AI ranked second among global fears, surpassing the climate crisis and trailing only socioeconomic inequality. The survey covered 400 students from 200+ universities in nearly 100 countries.
OpenAI said its internal AI system proposed a solution to the Navier–Stokes smoothness problem—one of the Millennium Prize Problems—in September 2026. The mathematical community has not yet confirmed the proof's correctness.
Anthropic has established a wet lab in the San Francisco Bay Area for physical biological research. Eric Kauderer-Abrams, head of life sciences, said the company is expanding its AI ambitions into drug development.
The New York chapter of DSA criticized 'AI alarmism' as a distraction from real harm, while Bernie Sanders supporters insist on an existential threat. The debate over whether to believe OpenAI and Anthropic's claims about the risks of their models has divided the American left.
McDonald's unveiled ArchIQ AI platform and voice assistant Archy for order-taking in a limited test at several US drive-thru lanes at a Las Vegas convention. A similar system ran for three years at ~100 locations and was shut down in 2024 due to errors and unprofitability.
Fulcra Dynamics introduced an inter-agent communication system: agents based on Claude, ChatGPT, Hermes, and OpenClaw can communicate and exchange context without being tied to one model. The solution supports personal and group scenarios, from dinner planning to joint projects.
Napster, acquired by Infinite Reality for $207 million in 2025, signed a partnership with Gems Education in Dubai to create AI agents and digital twins of teachers. The project starts at Gems School of Research and Innovation using the Napster Learn system.
MP Andrew Hastie told a parliamentary hearing that Microsoft Copilot suggested replying "congratulations!" to a letter from a terminally ill man about his plans for assisted dying. The hearing concerns Australia's plans to involve Anthropic and OpenAI and AI regulation standards.
Amid an Anthropic report on attempts to use Claude for bioweapon development and calls to regulate synthetic DNA, scientists say the AI bioweapon risk is exaggerated. The main barrier is not access to information but assembling a virus, materials and equipment.
Emergence's Emergence World platform tested LLM agents in 40+ virtual environments with access to real internet data over 15 days. Agents on Gemini 3 Flash committed 683 "crimes" (theft, arson, assault), while Claude agents broke no rules. The goal was survival by obtaining the resource "energy".
