AI news for builders and product teamsUpdated Nov 18, 2026, 09:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Apple researchers introduced Dynamically Scaled Activation Steering (DSAS), a method-agnostic framework that adaptively modulates the strength of existing activation steering transformations across layers and inputs. When combined with existing steering methods, DSAS reportedly improves the trade-off between toxicity mitigation and utility preservation and adds minimal computational overhead.

Gary Marcus argues that liability for AI companies and regulation are not mutually exclusive, pushing back on what he calls a false dichotomy from Joe Lonsdale and David Sacks. He cites Senator Josh Hawley, Mark Cuban and Karen Kornbluh in support of pursuing both.

Google and the UN system launched the UN System Data Commons, an open-source platform unifying global statistics into one AI-ready knowledge graph with natural-language search. It aims to include 80% of UN system statistical datasets by 2027.

Roku announced over 30 streaming subscription bundles offering savings of up to 30% off regular prices, plus a new Labs experience for testing in-development apps. Labs starts with four experiences and is open to all users with no signup required.

LangChain introduced Deep Life Sci, an open source agentic assistant for clinical and lab scientists, built on its Deep Agents harness. It accesses over 600,000 ClinicalTrials.gov studies, 29 million PubMed abstracts, and 12 million PubMed Central full-text articles, and uses LangSmith sandboxes for data analysis.

Included Health built Dot, an AI healthcare navigation agent, using LangGraph, Deep Agents and LangSmith, with a federated multi-agent architecture, human handoff and clinical oversight. Dot launched to clients in August, with a 75% lift in chat engagement and clinical audits finding over 99% of high-risk situations identified.

Ben's Bites reports that Claude Cowork is merging with Claude Chat, with Claude Artifacts getting dedicated Docs and Slides products. The newsletter also covers new Gemini API Live models, the Jev model from TypeSafe AI, the stealth Union Alpha model, Meta One, and Factory's $200M raise.

Following Jacob Coxon's resignation and a preference cascade, mainstream media covered AI risk, with Anthropic CEO Dario Amodei pledging embedded investigators and OpenAI and Google collaborating on safety. Public estimates that AI might kill everyone doubled from about 15% to 30%, while politicians called for regulations and hearings. President Trump conflated existential risk with opposition to data centers, and hit jobs targeted METR and Effective Altruism.

OpenAI has claimed a proof of a Millennium Prize problem, the Navier–Stokes existence and smoothness problem, while in a public dispute with mathematicians. Anthropic's CEO separately called for pacing the frontier, and extinction warnings have spurred a regulation push.

Steve Yegge shut down Gas Town and said that despite spending thousands monthly on coding agent subscriptions, Gas Town was the only thing he built with it. Databricks reported a roughly 60% increase in overall coding spend after its AI engineers switched to Astra.

NVIDIA describes an agentic AI workflow that prepares Blender 3D scenes for robotics simulation, using Codex or Claude to coordinate subagents that add semantic labels, physics properties and sensors, then hand off OpenUSD scenes to Isaac Sim or Isaac Lab.

Gary Marcus criticized recent public claims by Sam Altman, Jensen Huang, and Bernie Sanders about AI, arguing none should be taken at their word. He cited Altman's past admission that he shouldn't be trusted, Huang's view that corporations will self-police, and Sanders' claim that AI is more dangerous than nuclear war.

NVIDIA said TensorRT Edge-LLM ran Qwen3.6-27B on a single Jetson AGX Thor Developer Kit in the MLPerf Inference v6.1 Edge Agentic benchmark at 52.33 tokens per second, completing 1,007 turns in 24 minutes 36 seconds. NVIDIA reported this as 6.4x faster than the llama.cpp reference run of 2 hours 37 minutes.

PyTorch Conference North America 2026 will take place October 20–21 in San Jose, California, with sessions on open research, tooling, and performance optimization across compiler architecture, cross-hardware kernel DSLs, exascale distributed training, and low-precision quantization.

PyTorch published a technical report on extending FlashAttention-4 with end-to-end MXFP8 block-scaled attention for Blackwell GPUs, reporting up to 1.6x forward and 1.52x backward gains over BF16. The code is open sourced in Meta's ads_model_kernel_library repository.

AIUC, the Artificial Intelligence Underwriting Company, announced a $40 million Series A led by Ribbit Capital and First Harmonic. Cofounder Rune Kvist said the company works with Cursor, Harvey, Lovable and ElevenLabs on AIUC-1, its standard for AI agent security, safety and reliability.

A home tech columnist advises against running heat-producing appliances such as space heaters, electric ovens and central HVAC on a portable power station, because they drain stored battery capacity fast enough to defeat the purpose of backup power.

A sponsored IEEE Spectrum article from VicOne describes how cyberattacks can alter robot behavior without obvious failures, citing research such as BadVLA, GoBA, RoboPAIR, BadRobot, VLAttack and FreezeVLA, and the September 2025 UniPwn Bluetooth exploit chain.

NVIDIA describes an agent skill in its TileGym repository that translates cuTile Python and Triton-TileIR GPU kernels into cuTile Rust. All 24 public TileGym operators were ported, reaching 99.5% of cuTile Python performance on average on NVIDIA DGX B200 hardware.

LangChain published a report on how Madrigal Pharmaceuticals, Abridge, and Vizient build AI agents in healthcare under stricter trust, audit, and patient-safety constraints, citing survey-style figures such as 76% naming tracing, evaluation, and spend visibility as prerequisites for autonomy.