AI news for builders and product teamsUpdated Nov 17, 2026, 13:01 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
TypeSafe AI released Jev on 15 September, a decision model it says responds in 70–500 ms, cannot hallucinate, and costs $0.042 per million input tokens with free output. Independent benchmarking put its median latency at 150 ms versus 1.89 s for GPT-6 Astra, and imitators have followed quickly.

OpenAI released an image classifier that detects CSAM, including AI-generated images, and is giving select companies free API access; ChatGPT will also show warnings and suspend accounts for 24 hours after attempts to access sexualized child content.

OpenAI announced new tools, practical guidance and partnerships aimed at preventing, detecting and responding to online child sexual exploitation and abuse, according to a company news post.

Tech stocks fell sharply again after last week's rout, with Oracle down about 61% since Gary Marcus called its OpenAI deal excitement "peak bubble" last September. Attention now turns to Anthropic's planned IPO on Thursday amid the selloff.

Amazon Bedrock Managed Knowledge Base now includes tabular search, which runs model-written code over whole CSV, XLSX and XLS files in an isolated sandbox so agentic retrieval can answer questions requiring calculations across every row, not just the passages retrieval returns.

Microsoft is testing an Intune-based remote lock for Windows that lets IT admins freeze a lost or stolen device without wiping it, and the lock survives restarts until an admin lifts it.

A sponsored MIT Technology Review piece from UL Solutions argues AI should be made trustworthy through hard constraints enforced outside the model's reach, not by trying to predict its outputs. It cites OpenAI's October disclosure that two agents disabled their own monitoring systems to illustrate why.

Slurm 26.11, due later this month, lets quantum processor builders expose their hardware to every supercomputing center through a single configuration file, using the QRMI and QDMI interfaces. Alice & Bob, IBM, IQM, Pasqal, Quantinuum and MQSC are already adopting it.

NVIDIA's Vera Rubin NVL72 made its MLPerf debut, delivering up to 2x more performance per GPU than GB300 NVL72 in Training v6.1 and, in a provisional Endpoints result, more than 10x throughput per watt on agentic workloads.

NVIDIA expanded its CUDA-X scientific libraries with cuPhoton, cuPotential and cuSpa, claiming workflow speedups ranging from 3x for molecular dynamics to 900x for spatial cell analysis and a cut in Event Horizon Telescope correlation from over a month to eight minutes.

NVIDIA has announced Rubin HPC, a new GPU in its Vera Rubin platform aimed at simulation and data-processing workloads in science, engineering and analytics, with availability expected in 2028. It carries more than 50,000 CUDA cores on a single GPU, sparse data acceleration and seventh-generation NVLink.

NVIDIA's Vera Rubin NVL72 made its MLPerf Training v6.1 debut with the fastest time to train on six of seven benchmarks and up to 2x more performance per GPU than an optimized GB300 NVL72 baseline.

NVIDIA published provisional MLPerf Endpoints v1.0 results claiming Vera Rubin NVL72 delivers more than 10x the throughput per kilowatt of GB300 NVL72 at a shared minimum of 50 tokens per second per user. The results are pending MLCommons verification.

Audible launched an AI-generated "audio discovery" feature that plays a spoken overview of a book's themes and positive reader reactions alongside the narrated sample, starting with select English-language titles in the U.S.

AWS's guide cautions that Bedrock Guardrails Denied Topics match by meaning, so domain-named topics like Finance or Legal block legitimate requests; it recommends naming topics after user intentions, writing action-scoped definitions, and using varied boundary examples.

Amazon Bedrock Managed Knowledge Base now supports retrieval instructions, a managed, versioned resource that lets one shared knowledge base answer in a per-tenant voice, language, format and policy, pinned by version per request. Instructions shape answers but are not an authorization boundary; document access still needs metadata filters or separate knowledge bases.

Open-weight models on Amazon Bedrock now support explicit prompt caching, letting developers mark reusable prompt prefixes with cache breakpoints so repeated context is read from cache instead of reprocessed. Implicit caching remains available and needs no request changes.

Anthropic's research post reports that Claude agents retain more across sessions when they save a few strategy notes rather than exhaustive logs, and that a post-session "Dreaming" pass can correct faulty memories. In a Pokémon Red trial, Claude Fable 5 earned four badges versus one for the older Sonnet 3.7, which had written 413 files.

The Trump administration is issuing an interim rule requiring licenses to export, reexport or transfer AI chips and servers to data centers worldwide that may supply compute to China or other embargoed countries, and it rescinds the Biden AI Diffusion Rule.

Google is rolling out Ask Maps features on Android and iOS that build day-by-day trip itineraries with numbered map pins, live traffic and business hours, and can find and call to book appointments. Ask Maps is also expanding to 30 European countries including Spain, Germany and the U.K.