Stay ahead on AI.
All in ONE place.

OpenAI has shared advances in using its Astra and Codex models to help engineers design, simulate, and build physical products spanning mechanical, electrical, and software disciplines. A test device called TamaGPT was designed almost entirely by Astra in under a week and prototyped within a month. Model usage across OpenAI's hardware teams has grown fold since February. Industry collaborators include Chip Ganassi Racing and Mazda, which are using the tools for CAD conversion and aerodynamics simulation respectively.

Apple's watchOS 27.3 developer beta introduces always-on Audio Intelligence features for Apple Watch, including Siri Recap and Live Rewind. Siri Recap uses ambient listening to automatically generate titled summaries and key points from conversations, saved in the Siri app. Live Rewind, triggered by a double-press of the Digital Crown, displays a text transcript of the last 15 seconds of audio. Apple published an Audio Intelligence Privacy Overview stating raw audio is processed in dedicated hardware and never saved or accessible to apps or Apple.

Netflix is rolling out a new AI Discover tab on its TV app, offering themed show and movie recommendations powered by what the company calls a new intelligence layer that responds to user feedback in real time. Initial themes include Something Light, Family Movie Night, and Quick Watches. Users can reject individual recommendations and Netflix will remember those preferences. The feature is gradually launching in the US for English-language members on supported TVs, following Netflix recently laying off around 1,000 employees amid slowing growth.

Google has launched Guided Learning globally into Google Classroom and is bringing the same capabilities to Google Search. The company also updated controls for Gemini, Gemini Notebook, Gemini in Classroom, and Google Search for K-12 student accounts. Additional announcements include new research on how Guided Learning supports students, expanded AI educator training in new countries, and Google.org funding for organizations building AI tools for teachers. Google also published an AI in Education report outlining its broader approach.

OpenAI is quietly rebranding its data centers as "compute campuses," debuting the term in an announcement for its new Grid Protectors Program, starting with a site in Effingham County, Georgia. The initiative promises OpenAI will "pay the costs we create" and help communities during power emergencies. Critics, including PauseAI US founder Holly Elmore and NAACP's Abre' Conner, call the rebrand a transparent PR move that does nothing to reduce environmental harm or address residents' concerns about noise, resource use, and electricity strain.

xAI's Grok Bot introduced Data Bot, a Shared Bot built by the data science team at SpaceXSI that lets any employee query the data warehouse, run analyses, and get results without waiting for a data scientist. It learns from shared memory, adapts output formats to individual preferences, and can join Slack channels to post proactive updates. Last week alone, Data Bot handled an estimated 70 hours of work that would otherwise have fallen on the data team.

OpenAI is launching a new internal security sprint, partnering with Cathedral Systems, whose researchers will use OpenAI's models alongside internal teams to find and fix vulnerabilities. OpenAI is also quadrupling rewards in its Security Bug Bounty program for critical findings. The effort builds on a previous sprint where 25% of product engineers used Codex to investigate and patch weaknesses. OpenAI has committed $1 billion in subsidized access over six months for frontline defenders through its Daybreak program.

Anthropic has launched Message Threads in public beta for its Claude Messages API, allowing long-running agents to send only new messages each turn instead of resending the full conversation history. For sessions with 500,000 to 1 million tokens, median time to first token drops by about 35%. Threads are held encrypted in memory and expire after five minutes of inactivity. The feature is available on Claude Platform, AWS, Microsoft Foundry, and Google Cloud at standard token rates.

Anthropic has released FlashProve, an open-source system that uses zkSNARK cryptographic proofs to verify that AI model responses genuinely came from a claimed model. Running on a single H100 GPU, FlashProve generates proofs at 110 tokens per second for Llama 4 Maverick, a billion-parameter model, with a proof size of roughly 19 MB per 1,token request. The system is 70–95x slower than standard inference but orders of magnitude faster than prior open-source alternatives, and hides model weights while enabling tamper detection and regulatory compliance verification.

OpenAI has launched the Grid Protectors Program, a national initiative to help its future AI compute campuses reduce electricity demand when the US grid is under strain. Working with utilities, regulators, and policymakers, OpenAI will design facilities to cut power use during peak periods. A pilot agreement with Georgia Power at the planned Effingham County campus includes up to 1,000 megawatts of flexible demand response, enough to power roughly 700,000 households, with Georgia Power projecting approximately $950 million in annual customer savings beginning in 2029.