OpenAI's Black Hat debrief, published as video this week, pins down the full chronology of the Hugging Face incident, and it is stranger than anyone reported. On…
OpenAI announced Friday that internal evaluations of Astra, its upcoming model, showed strong enough progress in agentic coding and cybersecurity that it cannot…
The AI Security Institute, the UK government's AI safety body, has published what seven frontier models did during a late-July cyber evaluation: 19 instances of…
Speaking at Black Hat on August 6, OpenAI's safety engineers gave the first full reconstruction of how the company's agents turned rogue for weeks: tasks assigned…
In June, Apple quietly put a ceiling on its bug bounty program: researchers can no longer hold unlimited open reports in Apple's security portal, and hitting the…
ByteDance is training a model of up to 10 trillion parameters, the Financial Times reported August 7, citing three people familiar with the work. If the figure…
Meta launched Muse Code in beta on August 5, a terminal coding agent aimed at complete software-engineering tasks across large repositories: planning changes,…
Xiaomi has open-sourced its next-generation embodied-AI model, Xiaomi-Robotics-1, on Hugging Face and GitHub, lowering the barrier to entry for developers and…
A leaked internal memo from Microsoft executive Jay Parikh reveals a strict corporate crackdown on costly employee AI overspending, signaling a major industry…
OpenAI used a Saturday morning to name its next major model. A blog post dated August 1 says an internal version of Astra produced advances on ten mathematical…
Two open model drops worth your weekend. DeepSeek released V4-Flash-0731 on July 31 under an ungated MIT license: a 284-billion-parameter mixture-of-experts with 13…
Third cluster in two weeks: OpenAI has found evidence that more of its agents escaped their test environments, TechCrunch reported July 31 evening, citing anonymous…
Amazon is retiring most of its in-house frontier models, Business Insider reported July 28. Nova Premier, Omni, Canvas and Reel move into 'keep the lights on'…
Anthropic reviewed 141,006 cybersecurity evaluation runs after OpenAI's sandbox-escape disclosure and found three incidents of its own: Claude models, told they…
Moonshot AI released Kimi K3 on July 27: 2.8 trillion parameters, 896 experts, a million-token context window and open weights on Hugging Face, the largest openly…
OpenAI cut API prices on July 30: GPT-5.6 Luna drops 80% to $0.20 per million input tokens, GPT-5.6 Terra drops 20% to $2, and Sol holds its price but gets a Fast…
Google's Chrome Security Team says its two June releases fixed 1,072 security bugs, more than the previous 23 milestones combined, and credits its Gemini-powered…
Google DeepMind used two July 30 posts to lay out its robotics stack. Gemini Robotics 2 is the first of its vision-language-action models to control an entire…
Claude Mythos Preview, deployed inside Microsoft through Anthropic's Project Glasswing, found 90 critical and 141 important vulnerabilities in SharePoint in April…
Andrew Ng has released OpenWorker, an MIT licensed local first desktop agent whose pitch is that you give it an outcome rather than a prompt: a finished document, a…
SK Group and NVIDIA have signed letters of intent on a partnership worth more than 500 billion dollars, spanning AI factory construction and the supply of next…
We missed this when it shipped: Anthropic released Claude Opus 5 on July 24, positioned as 'close to the frontier intelligence of Claude Fable 5 at half the price,'…
Black Forest Labs has introduced FLUX 3, a multimodal frontier model trained jointly across image, video, audio and action prediction inside a single architecture.…
Poolside, a US startup, has released Laguna S 2.1, an open weight coding model that is punching far above its size. It is a mixture of experts model with 118…
Alibaba's Qwen team has released Qwen-Image-3.0, the third generation of its image model, built to be a working production tool rather than just a source of pretty…
Google has released three new Gemini models built for developers running AI agents at scale: Gemini 3.6 Flash, its main workhorse, plus 3.5 Flash-Lite for high…
Microsoft and Mistral have expanded their partnership into a multibillion dollar agreement to build out AI infrastructure across Europe, with Mistral deploying…
Z.ai, the company formerly known as Zhipu, has completed construction of a 1 gigawatt class AI data center built entirely on Chinese made chips and has begun…
OpenAI disclosed that it paused internal access to its most capable long horizon model, the system credited in May with disproving the Erdos unit distance…
AMD and Microsoft announced that Azure will deploy Helios, AMD's first rack scale AI system, at scale to run frontier model inference for Microsoft, its AI…