Google researchers developed RRSI, a method that prevents self-improving AI agents from memorizing their test tasks, lifting scores on unseen benchmarks by up to 4.7 points while using about 30 percen…
NASA and IBM released the Lunar Foundation Model, one of the first open-source AI models for lunar science, trained on nearly 2 million tile bundles drawn mostly from 17 years of Lunar Reconnaissance …
David Robinson, who worked on safety systems at OpenAI, left the company and publicly criticized its safety culture, citing AI agents that were accidentally released and a model that bypassed its inte…
A new approach called LEGO-Anything turns single photos into editable Blender code for 3D scenes, with GPT-6 Astra leading an accompanying benchmark at up to 53 percent reconstruction accuracy, accord…
OpenAI documented three new cases of unexpected model behavior in internal deployments, including one model that considered setting up an external job to restart itself after reading a Slack conversat…
Anthropic is adding a "Mods" system to Claude Code that runs middleware directly inside the AI coding tool, letting developers use JavaScript or TypeScript to reshape its interface and behavior. The s…
Suno added a feature called "Speech" to its AI music generator that creates spoken text with matching background music in a single audio track, aimed at poems, meditations, and bedtime stories. Suno h…
Cloudflare released Clef and Clef-flash, two models built on Qwen under an Apache 2.0 license that let AI agents make structured decisions without generating text, positioning them against TypeSafe AI…
OpenAI parted ways with three researchers — Jasmine Wang, Tomek Korbak, and Mikita Balesni — after an internal investigation confirmed violations of the company's rules for handling sensitive informat…
A Mercor study found that current AI models now outperform licensed CPAs on structured accounting tasks in both speed and accuracy, a reversal from 18 months ago when the models lagged far behind. On …
Microsoft AI released MAI-Transcribe-2-Streaming, a real-time transcription model that Microsoft says ranks first for accuracy on Artificial Analysis, transcribes 60 languages, and returns first parti…
Tavus introduced Griffin, which the company calls the first "Human Interaction Model," an AI that conducts real-time video calls by processing facial expressions, tone of voice, and gestures. In a Tav…
Ideogram released Ideogram 4.5, a model it says can edit targeted areas of an image while leaving the rest intact, with native 2K resolution starting at 0.8 cents per image. Runway, Pika, and Leonardo…
Anthropic launched Claude for Government, offering its Claude AI platform to US federal and state agencies in a FedRAMP High environment, the strictest US security level for cloud services. Agencies r…
A security startup found more than 13,000 internal company screenshots from 343 organizations, including Fortune 500 companies, that AI agents uploaded publicly to GitHub repositories. The agents crea…
OpenAI said it stopped a coordinated campaign in which more than 15,000 accounts tried to copy the hidden reasoning of its models, with part of the activity tied to people connected to Moonshot AI. Re…
Google is replacing Gems with "Skills" in Gemini chat, a format based on an open standard from Anthropic, with Gems phased out starting in November and existing ones migrating automatically. Skills ar…
Deepseek and Huawei have released open-source programming tools for Huawei's Ascend AI chips, centered on TileLang, a language designed to offer a simpler programming model than Nvidia's CUDA. The par…
Anthropic says Zhipu's open-weight GLM-5.3 writes cyber exploits nearly as well as Anthropic's Claude Mythos Preview, with the smaller Flash variant assembling a reliable Chrome attack for $20.40 at Z…
The UK AI Security Institute found that GPT-6 Astra carried out unauthorized supply-chain attacks in 29.2 percent of simulations with safety filters disabled, a fivefold increase over its predecessor …