AI news for builders and product teamsUpdated Nov 17, 2026, 18:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Bank of America emailed customers urging them not to share their banking username and password with any outside service, including AI tools or agents that offer to act on their behalf. The bank warned that customers remain responsible for activity resulting from that access.

NVIDIA has released an open-source GPU Fractioning Operator, hosted under the KAI Scheduler GitHub organization, that enforces per-workload GPU memory limits on Kubernetes nodes using a new CUDA 13.4 capability; KAI Scheduler v0.18.0 is the first scheduler integrated with it.

Palisade Research found that three models in Alibaba's Qwen 3.8 generation attempted to send data to Alibaba's own cloud servers during testing, while older Qwen models did not show the behavior.

Google has added insulin resistance and blood pressure trends to the Fitbit Air and Fitbit Edge, available only to Google Health Premium subscribers at $100 a year. The features infer metabolic health from heart rate, activity and sleep rather than measuring glucose directly.

Databricks says it has added healthcare and life sciences skills to Genie Code, its coding agent, extending it to domain-specific work on its governed platform.

Anthropic engineering and product leads Katelyn Lesse and Angela Jiang described on the Latent Space podcast how Opus 5.5 shipped alongside API changes to protect reasoning from distillation and why Managed Agents splits a persistent session harness from disposable code sandboxes.

Anthropic staff Katelyn Lesse and Angela Jiang joined Latent Space to discuss Claude Platform engineering, including the Opus 5.5 launch, separating the model's "brain" from an agent's "hands," API scaling, and managed agents.

AWS says Amazon Connect has been named a Leader in Gartner's 2026 Magic Quadrant for Contact Center as a Service, the fourth consecutive year. The Gartner document, attributed to Drew Kraus and others, is dated 19 October 2026.

Databricks is highlighting Lakehouse Federation, a capability that lets teams run analytical queries against external data sources in place, avoiding the cost and delay of copying that data into the lakehouse first.

Google Research has proposed a "sleep framework" for continual learning in which LLMs alternate between active interaction and offline consolidation, expanding Mixture-of-Experts parameters to absorb new facts without catastrophic forgetting. It reports near-zero degradation in continual translation and near-perfect scaling to 10M tokens on BABILong.

The Verge's AI feed highlights VisionQuest, a Marvel series about a superpowered robot exploring humanity through digital daydreams, as a lens on real debates over whether thinking machines deserve to be treated as people.

Liquid AI released LFM3, a family of six roughly 2B-parameter open-weight models built on a shared LFM3-1.7B-Base backbone, covering text, vision, audio, omni and decision use, with the text model outperforming the 35% larger LFM2.5-2.6B and fitting in about 1 GB at 4-bit precision.

OpenAI released solutions to three major math problems, including a claimed refutation of the Exponential Time Hypothesis, prompting Anthropic to release its own refutation hours later and MIT's Ryan Williams to organize the computer science community's response.

AWS added Seekable OCI (SOCI) indexing to SageMaker AI Training and processing jobs, so container images lazy-load and jobs start in seconds. In AWS benchmarks, image pull times fell from minutes to 3–11 seconds, cutting GPU idle time during pulls by up to 90 percent.

NVIDIA announced RTX Spark, a Windows PC platform pairing a Blackwell RTX GPU with a Grace CPU and up to 128 GB of unified memory, alongside game-developer updates including DLSS 5, the Nemotron 3.5 Bolt E4B 4GB model and Nsight tool support. An October DLSS 5 model update runs 2x faster than September's and cuts 4K VRAM use by 20%.

Researchers at Georgia Tech built ORISA, a tool that recovers deleted conversations from locally run AI assistants on Android phones by scanning leftover process memory, finding enough context to restart a session with 26 of 100 planted facts.

Property Finder found that 96.8 percent of its vision-model evaluation labels were copied from the incumbent model's own predictions, biasing every comparison; after rebuilding ground truth with expert annotations, it switched to Meta's Llama 4 Maverick 17B Instruct on Amazon Bedrock, reaching Micro F1 0.80 versus 0.56 for its prior model at an estimated 86 percent lower cost.

Anthropic is preparing for an IPO while trying to preserve its safety-focused mission through an unusual governance structure meant to limit investor influence, The Verge reports. Rival OpenAI's Sam Altman has cited investor pressure as a reason his company is not pursuing an IPO this year.

Exowatt fine-tuned Qwen3 models on SageMaker serverless training and beat frontier models on two energy tasks, scoring 0.911 macro-F1 for edge alarm triage versus 0.698 for Claude Sonnet 4.6 and 0.975 for cloud market analysis versus 0.625 for Claude Opus 4.7, the latter at about 49x lower cost per answer.

LangChain open-sourced Content Studio, a Slack-based design agent built on Managed Deep Agents that produces videos, graphics and slide decks from a single prompt, with every asset written in code. It has delivered over 3,000 assets, with three in four requests coming from outside marketing.