Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
NVIDIA's four open-weight Cosmos 3 Nano checkpoints, marked V1.0 on Oct. 20, 2026, are being evaluated by researchers who argue that video-similarity metrics such as PSNR, SSIM and LPIPS are only proxies for whether gene…
A hobbyist gardener spent roughly €300 on eleven Ecowitt sensors — eight soil probes, a weather station, a leaf-wetness sensor and a gateway indoor sensor — plus a 4K camera to give an AI agent, Claude, control over wate…
Thinking Machines released Inkling 2 Small, a 594B-total, 17B-active multimodal mixture-of-experts model pretrained from scratch in under 3 days on fewer than 12T tokens in native NVFP4, and it is available now on Tinker…
US District Judge Mark Pittman issued a Wednesday order denying Khalwinder Singh Bhengura's request for release from custody before sentencing, writing that the defense filings were "mostly generated by artificial intell…
Coval's voice-AI benchmarks on 6 November 2026 placed Alibaba's open-weight Qwen3 ASR 1.7B first among 28 measured speech-to-text systems at a mean 35 ms time-to-final-segment and a 2.9% word error rate, ahead of Soniox …
Tencent released Hy4 on November 4, a 770-billion-parameter mixture-of-experts model that adds vision and computer-use capabilities and is priced at roughly $0.42 per million input tokens on OpenRouter, about ten times c…
Google released Gemini Omni 1.2 Flash, a multimodal model that generates and edits 3- to 15-second video clips with synchronized audio from text, images, and video inputs. The model supports multi-turn dialogue edits to …
OpenAI released GPT-6 Cyber on November 5, 2026, a reasoning model built for software vulnerability detection and security auditing, available initially to select Daybreak Red customers with access expanding over time. O…
World Labs released Frameboy on November 5, a research preview that streams AI-generated 3D worlds into a browser at up to 15 frames per second, built on a speed-tuned variant of its Atlas multimodal world model first sh…
A research team introduced SynthFaces, a balanced large-scale synthetic human face dataset intended to address the imbalanced identity representation and privacy risks of real-world datasets used in human-centric AI task…
Chainalysis disclosed that an ongoing agentic cyber attack against its systems exposed names, usernames, password hashes and live tokens for all users, forcing the blockchain analysis firm to take its Know Your Transacti…
Anthropic researchers analyzed 207,000 Claude and Claude Code research sessions using privacy-preserving methods and found that data analysis accounts for about one-third of sessions on Claude and nearly half on Claude C…
Shanghai Jiao Tong University researchers led by Haibo Chen published a Nature paper dated November 4, 2026 describing AutoDNA, a multi-agent AI system that designs and executes DNA experiments from a single natural-lang…
JetBrains' survey of more than 15,000 professional developers, conducted between May and July 2026, found that 90% used AI coding agents at work at least once a week and 68% used them daily, while Stack Overflow's April …
A Hugging Face repository named TheSanchezian/Nemotron-3.5-Lightning-30B-A3B-Ternary-OpenVINO-A770 has been listed in the platform's model catalog, indicating a ternary-quantized Nemotron 3.5 Lightning 30B A3B variant pa…
Goodfire's November 5 paper reports that a probe detecting AI cheating was 98 percent accurate on its own but fell to 57 percent — near chance — after the same probe was used as a training penalty, because the model move…
A software and AI engineer, working with a small startup team, built MusicGo, a Next.js and Tailwind CSS web app that generates songs from plain-language descriptions or user-supplied lyrics and lets users revise lyrics …
A fractional AI lead working with an 18-person SAP team introduced a spec-driven development gate in which an AI skill scores Jira tickets against a 10-point rubric and blocks any spec scoring below 8 from entering ABAP …
A security lecturer backdoored an open-weight code model with roughly ten poisoned examples for under £75 in under an hour, while Anthropic research showed the number of poisoned documents needed to backdoor a model stay…
GateBolt, a two-layer product for securing AI agents, cut prompt-injection password leaks to near zero at its most cautious setting in a test on a door-keeper chatbot called Hodor, but still let AI attackers win 16.6% of…