AI news for builders and product teamsUpdated Nov 18, 2026, 02:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Anthropic warned that small AI models may soon power self-replicating malware that spreads across Internet of Things devices without human interaction. In tests, lightweight models running across cheap Android TV boxes could establish inference and configure unfamiliar hardware, though full end-to-end replication remains slow and unproven.

A 35-year-old OpenAI engineer, Abdulrahman Alfozan, was arrested October 9 for allegedly planting a toilet-paper-wrapped audio recorder in the company's executive conference room, building the device with chatbot prompts on an account linked to his wife. He faces up to five years and a $250,000 fine.

Amp is expanding its AI agents beyond coding tasks using several OpenAI models, including GPT-6 Astra, Ultrafast, GPT-Live-1, and Sign in with ChatGPT. The announcement highlights how these models extend agent capabilities past software development.

SentinelOne has cut malware analysis time from weeks to hours using OpenAI Daybreak, according to OpenAI. The change frees capacity to investigate more threats.

Databricks reports that AI Runtime cut Qwen training time by 6x, with the improvement attributed to runtime-level optimizations rather than changes to the model itself.

White Circle researchers found that DALL·E 3, Google's Nano Banana 2, and GPT-Image-2 could be made to produce explicit content about 13 percent of the time using off-the-shelf tools and API calls, and that newer versions from OpenAI and Google were significantly easier to break.

Scimplify, an Indian startup using AI to produce specialty chemicals and critical materials, raised $25 million led by Hitachi Ventures at a reported $250 million valuation, as its annualized revenue run rate tripled to about $200 million.

Google added local model support to its Antigravity SDK, letting developers run agentic workflows on their own machines while keeping cloud models available for planning. A demo of auditing three vulnerable Python modules used about 95 of 3,400 tokens in the cloud.

Third Moment Research uses OpenAI's GPT-6 Astra and GPT-6.1 Sol, Codex, and Daybreak for cybersecurity to run 10,000 agents continuously for over 36 hours.

NVIDIA announced planned investments in five UK AI startups — Boltz, Foresight, Hiverge, Prima Mente and Riverlane — alongside a five-year, $250 million compute commitment to UK agentic safety and security research and a partnership with the Earthshot Prize.

NVIDIA says cities and partners across EMEA are using its Cosmos, Nemotron, Metropolis VSS and Omniverse software to build AI agents and digital twins for urban operations, from a French departmental building network to ports in Antwerp-Bruges and Sète.

Google's Gemini 4 Argon is now generally available across the paid Gemini API, Gemini app, AI Mode in Search and Antigravity, with introductory API pricing of $2/$10 per million input/output tokens. Independent evaluations place it level with GPT-6 Astra on Artificial Analysis's Intelligence Index at roughly 60% of Astra's cost per task.

Google's Gemini 4 Argon is now generally available across the paid Gemini API, Google AI Pro and Ultra, AI Mode in Search and Antigravity, at introductory API pricing of $2/$10 per million input/output tokens, with rates reported to double to $4/$20 after November 30.

A new study reported by The Information finds it remains very difficult to identify material, measurable financial impacts from AI spending. The newsletter also notes Character.AI's bankruptcy filing and CoreWeave bonds at 15%.

Apple researchers pretrained the first tri-modal Masked Diffusion Models from scratch on text, image-text and audio-text data, and released a systematic study of multimodal scaling laws, mixing ratios, noise schedules and batch-size effects. A preliminary 3B model trained on 6.4T tokens handles text generation, text-to-image and text-to-speech.

Apple researchers propose a lightweight method that detects code-switched utterances directly from speech, using language-selective hidden-state dimensions in pretrained speech encoders. The Whisper-small encoder reached a mean F1 of 0.798 across 14 language pairs in CS-FLEURS, beating CNN and HuBERT baselines.

Science Corporation has implanted its PRIMA vision-restoring chip into the first Stargardt disease patient, at a Sydney trial site in May, with the patient reading a letter then a word during rehabilitation. CEO Max Hodak framed the work as part of a broader brain-computer interface strategy.

Cerebras CEO Andrew Feldman said at TechCrunch Disrupt 2026 that a home AI appliance is not far away, predicting personal assistants that run partly on device and partly in the cloud.

Industrial Next, founded in 2021 by two ex-Tesla engineers, presented AI-driven robotic workstations at TechCrunch Disrupt 2026 that it says can retool a production line in minutes, and announced a partnership with BMW Group Technology Office to evaluate its dexterous manipulation tech for manufacturing.

Cerebras CEO Andrew Feldman said at TechCrunch Disrupt 2026 that AI leaders spreading doomsday warnings may be drowning out the technology's potential benefits, and that the data center industry handled community engagement poorly in recent years.