AI news for builders and product teamsUpdated Nov 17, 2026, 20:01 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Google, Temasek and Bain & Company project Southeast Asia's digital economy will reach $368 billion in GMV in 2026, up 17% year on year, and top $1 trillion by 2035, driven by generative AI used by 243 million people in the region.

TornadoVM 7.3 replaces JNI with the Foreign Function & Memory API and drops the experimental JVMCI dependency, giving one compiler-metadata path across JDK 21, 25, 26 and 27. The release adds an optimized CUDA backend with cuTile and cuBLAS/CUTLASS calls from TaskGraphs, plus multi-GPU execution.

TEAL, a pixel-level active learning framework from University of Waterloo researchers, picks diverse tokens directly on a frozen ViT's token lattice and beats the strongest baseline by up to +24.36 mIoU with one labeled pixel per image.

Private keys for eight eMudhra intermediate CA certificates were compromised, and because they chain to roots trusted by Mozilla, Microsoft, Apple and Chrome programs, software bundling trust lists must be updated. Python tools pip, uv and certifi have already shipped refreshed certificate bundles.

US District Judge Zachary Bluestone dismissed Ann Altman's lawsuit alleging childhood sexual abuse by her brother, OpenAI CEO Sam Altman, after she missed an independent medical examination and other deadlines over the past year despite warnings.

Anthropic's federal affairs lead Anthony Cimino publicly defended the company's leaked draft AI safety bill, saying it was shared with a small number of congressional offices and that its preemption and model-blocking language will be revised. Critic Zvi Mowshowitz argues the response misses the central problem: the draft itself is too weak for the current moment.

Anthropic is circulating a draft federal AI safety bill that would let a judge, not a federal agency, halt a model's release when it poses a foreseeable, imminent risk of killing more than 50 people or causing over $1 billion in damage.

i14 and AWS trained and served the Laplace 3B linear attention model on a single Trainium2 instance using native PyTorch via TorchNeuron Native, reaching about 120,000 training tokens per second and 11,544 output tokens per second across 1,280 concurrent sequences.

SpaceX reported third-quarter revenue of $14.6 billion and net income of $4.4 billion, a swing from a $1 billion loss a year earlier, with its AI unit generating $7.5 billion, or 52% of income.

OpenAI acquired Orchestra Bio, a four-year-old AI biotech software startup, for undisclosed terms. Orchestra's team and technology will work on OpenAI's Rosalind life sciences model and the Rosalind Workbench tool, which already counts Moderna and Novo Nordisk as customers.

Meta removed the word "train" and the phrase "any content" from the commercial terms of service taking effect October 30, after legal experts said the original language let it scrape sites linked from business pages. Meta did not notify users of the revision, and the experts who raised the alarm say the new wording still permits too much.

Google is rolling out Gemini's Personal Intelligence to Android Auto starting now, with vehicles running Google built-in to follow over the next few months. The opt-in feature lets Gemini draw on connected apps like Gmail, Calendar and Photos to answer requests in the car.

Google is rolling out Personal Intelligence to Gemini on Android Auto, letting the assistant pull context from Gmail, Calendar, Photos and other Google services to answer in-car questions. It is opt-in and off by default, and users can toggle access per app or service.

Simon Willison built Battle Drones, a browser-based real-time strategy game in the style of Command & Conquer, across 12 major iterations and 321 commits largely written by coding agents, including a multiplayer mode where two nearby devices synchronize game state over audio.

SF Compute has begun moving GPU capacity onto its order book, with 5% of total capacity tradeable so far, as CEO Evan Conrad described at AI Engineer New York. The company now runs much of a cloud's operating stack itself so buyers receive usable compute rather than a cash payout.

SF Compute has moved about 5% of its total capacity onto its trading order book as it shifts from reselling clusters to operating cloud-grade infrastructure itself, CEO Evan Conrad said at AI Engineer New York's new Compute Trading track.

Elevenlabs has released two speech-to-speech models, Eleven Realtime v1 Max and v1 Mini; v1 Max scores 98.8% on the Big Bench Audio benchmark with a 0.57-second response time.

Google released TorchTPU, an open-source PyTorch version that lets AI apps run on Google's TPUs with just a configuration change, joining Amazon's similar TorchNeuron effort to reduce Nvidia's dominance in AI hardware.

OpenAI is rolling out Locked Chats, letting users hide selected ChatGPT conversations behind a four-digit PIN or Face ID/Touch ID, separate from regular history and search. The feature reaches most ChatGPT plans except Healthcare, and OpenAI says it is not end-to-end encryption.

NVIDIA released Cosmos3-H-Dreams, a generative surgical simulator trained on clinical cholecystectomy recordings from the CMR Surgical Versius platform that runs interactively on a single RTX PRO 6000 GPU. Two checkpoints trade visual quality for inference cost via four-step and two-step denoising.