18–24 October 2026
The week's stories that will still matter next week, with the noise taken out.
SpaceX’s AI revenues surge
SpaceX said its Super Intelligence unit drove $7.6B of $14.6B Q3 revenue, outselling rockets and Starlink.
Agents get safer guardrails
Microsoft launched Guardian Agents to monitor autonomous tools, saying frontier agents can’t be fully trusted.
Models & labsOpenAI's Decisions API is generally available
Built on GPT-6 Luna, the Decisions API answers bounded questions from text, images or JSON.
Video generationFLUX 3 Max makes a 10-second clip in about 11 seconds
fal and Black Forest Labs launched FLUX 3 Max, a post-trained video model.
Story of the week
The single most consequential story of the week.

- Ban effective date
- Since July
- Order unveiled
- 19 Oct
US orders DeepSeek off civilian government systems
The US Department of Homeland Security told civilian agencies to remove DeepSeek and bar technology built with its models. The order, in force since July and made public on 19 October, immediately reshapes which AI tools federal teams can use.
Why it's the story of the weekA widely used Chinese model is now off-limits across US civilian government, setting a precedent other sectors may follow.
Models & labs
New models, the labs behind them, and the rules they answer to.
OpenAI's Decisions API is generally available
Built on GPT-6 Luna, the Decisions API answers bounded questions from text, images or JSON. OpenAI says it is up to 10x faster than Luna via Responses, and one request can hold 400 questions.
Why it mattersRouting and scoring can run in-line without paying for a full generation.
OpenAI ↗GPT-6.1 Sol lifts COBOL migration from 5.4% to 86.6%
OpenAI reported GPT-6.1 Sol reached 86.6% on its COBOL migration test at low effort, versus 5.4% for GPT-5.6 Sol, at 39% lower cost. At higher effort it reached 97.9%.
Why it mattersBanks and governments still rely on COBOL; safer, cheaper conversion widens what can be modernized.
OpenAI ↗Agents & coding
AI that acts on its own, and the tools developers build with.
Microsoft ships monitors for agents it says cannot be trusted
Microsoft launched Guardian Agents under Project Perception to watch company-run agents. Satya Nadella urged separating intelligence providers from authority to act.
Why it mattersAutonomous tools need oversight akin to insider-threat monitoring.
TechCrunch ↗Walmart opens a checkout built for agents
Walmart’s Agent Gateway lets approved agents search Walmart and Sam’s Club and complete purchases without seeing payment details. Standard returns still apply.
Why it mattersWhile some sites block agents, a top retailer is welcoming them with guardrails.
TechCrunch ↗Video generation
The models and tools that make video.
FLUX 3 Max makes a 10-second clip in about 11 seconds
fal and Black Forest Labs launched FLUX 3 Max, a post-trained video model. On a preliminary arena it scored 1,139 Elo from 4,868 votes, and fal cites 10.95 s for a 10 s clip, versus 277.8 s for Seedance 2.5.
Why it mattersNear real-time generation lets creators iterate while the idea is fresh.
PR Newswire ↗EdenLive renders live game frames on one RTX 5090
VAST released EdenLive code and 5B/14B weights. It says the 5B model can render 720p at 24+ fps on a single RTX 5090.
Why it mattersA live renderer on one consumer GPU could change how games draw scenes.
DigIntoAI ↗Adobe's Frame.io search gets about five times faster on Nvidia chips
Adobe is running Frame.io’s video search on Nvidia GPUs for faster embeddings, throughput and latency. Editors can search by concept, action or setting.
Why it mattersFaster semantic search helps find moments across years of footage.
Nvidia ↗Creator tools
The apps and platforms people make video with.
Cursor adds cloud agent canvases and env API
Cursor’s cloud agents can now build shareable interactive canvases. A beta Environments API manages runtimes with support for up to 300 repos and faster loads.
Why it mattersCoding agents gain richer UIs and scalable environments for real projects.
cursor.com ↗TorchTPU brings native PyTorch to TPUs
Google and Meta open-sourced TorchTPU, a PyTorch backend that runs existing code on Cloud TPUs via torch.device("tpu"), including eager mode and torch.compile.
Why it mattersTeams can shift PyTorch workloads to TPUs without rewriting into another stack.
Google Cloud ↗Media & entertainment
AI in the films, shows, micro-dramas, music and ads people watch, the studios and platforms making them, and the rules for AI-made content.
Disney agrees to hire most of Character.AI's staff
TechCrunch reports Disney will offer jobs to at least 60 Character.AI employees, about $60M a year in total comp. Most accepted; another buyer could still outbid it.
Why it mattersA bankrupt chatbot startup is sending much of its talent to a studio.
TechCrunch ↗Spotify says the remix tool has support beyond four partners
Spotify told Billboard its remix tool has backing beyond Universal, Merlin, Kobalt and Believe, with artist consent required. It also pulled 50M+ songs from recommendations as AI spam.
Why it mattersMajor platforms are policing AI spam while testing creator tools.
Billboard ↗A White House column photo is flagged as AI-edited
A Truth Social before-and-after image was marked by Google SynthID as edited with OpenAI. The restored side shows mismatched steps; other photos were not flagged.
Why it mattersOfficial-looking posts with AI marks complicate public trust.
The Verge ↗Chips & compute
Chips, data centres and the power behind them.
SpaceX's AI business outsells rockets and Starlink
SpaceX reported $14.6B in Q3 revenue, up 161% YoY, driven by its Super Intelligence segment. The unit brought in $7.6B from cloud contracts, Cursor and Grok.
Why it mattersA rocket company now earns more from AI than from launches and broadband.
The Verge ↗Microsoft drops an Essex AI datacentre over power
Microsoft exited as anchor tenant of Nscale’s planned Loughton site after the grid couldn’t supply power until the early-to-mid 2030s.
Why it mattersPower access, not chips, is stopping some AI campuses.
The Guardian ↗Policy & safety
Laws, lawsuits and safety incidents.
German courts treat Google's AI answers as its own
Aachen’s court barred Google from repeating false AI Mode claims about a doctor, treating the text as Google’s own with fines up to €250,000 per breach.
Why it mattersIf AI answers are the platform’s words, liability risks rise sharply.
The Decoder ↗Nvidia, Anthropic, and OpenAI set lobbying records
From July to September, Nvidia spent $2.2M, Anthropic $2.1M, and OpenAI $1.5M on US lobbying, each a record quarter for the company.
Why it mattersPolicy stakes are high enough to drive record corporate lobbying.
The Verge ↗Robotics
Robots, self-driving and AI in the physical world.
Tesla starts production-quality Optimus in Fremont
Tesla said the first production-quality Optimus robots are coming off a line that used to build cars. The company says the process is fit for a commercial product.
Why it mattersA major EV maker is shifting a car line to humanoid robots.
TechCrunch ↗Waymo is reported to be aiming at 20,000 robotaxis
Waymo plans to scale from ~4,000 to 8,000 cars by end-2026 and 20,000 by end-2027 after a $5B loan, Semafor reported. It will also test around Madrid, aiming at first riders in 2027.
Why it mattersLarge-scale robotaxi fleets are moving from slideware to concrete targets.
The Next Web ↗Digit 5 will standardise on Nvidia IGX Thor
Agility Robotics said GA Digit 5 will use Nvidia IGX Thor exclusively and expand Halos models for human detection. Early access is planned for H1 2027.
Why it mattersNvidia becomes the exclusive compute supplier for that production humanoid.
Agility Robotics ↗Science & health
AI in biology, medicine, climate and maths.
Microsoft's diagnostic panel beat generalist doctors in simulation
A Science study reports MAI-DxO, a panel of AI agents, correctly diagnosed 83% of complex NEJM cases versus 36% for general internists and primary-care doctors. Gains were strongest on complex cases.
Why it mattersAgent panels show promise on difficult diagnoses, a key pressure point in healthcare.
TechCrunch ↗AI can predict when climate systems will reach a tipping point
Caltech, CU Boulder and Nvidia trained an AI that forecast a climate threshold more than two years ahead with an 11-day error in a test.
Why it mattersEarlier tipping-point warnings could buy time for policy and adaptation.
Earth.com ↗Open source
Open-weight models and community releases.
DeepSeek open-sources the engine behind its API
DeepSeek released DeepInfer under Apache 2.0 with support for Nvidia SM100 and Huawei Ascend 950. The GitHub drop omits the cache and quickly drew thousands of stars.
Why it mattersOpening a production inference stack spreads the know-how beyond one lab.
The Next Web ↗Flint is a laptop coding model under Apache 2.0
Recursive released Flint, a 40B MoE coding model with 2.5B active, trained with heavy automation. It led sub-50B models on SWE-Bench Verified.
Why it mattersA strong local coder under a permissive license widens offline options.
TechCrunch ↗Anthropic gives Agent Skills to the Linux Foundation
Anthropic donated its Agent Skills folder/SKILL.md standard to the Agentic AI Foundation. More than 70 platforms use it, with about 1M skills and nearly 280M installs counted.
Why it mattersA common skill format makes it easier to share and reuse agent abilities.
Anthropic ↗Paper of the week
One research result, explained in plain words.
Microsoft's diagnostic panel beat generalist doctors in simulation
Mustafa Suleyman’s team; Dominic King quoted
- What they tried
- Coordinated several AI agents (MAI-DxO) to question patients, order tests and debate diagnoses, measuring accuracy and estimated cost across complex NEJM cases and emergency department cases.
- What they found
- On complex cases the system was correct 83% of the time vs. 36% for general internists and primary-care doctors; gains were strongest on complex cases and with less capable base models, and smaller on emergency cases.
- Why it matters
- Agent panels that coordinate steps and evidence can outperform generalist doctors on hard diagnostic problems in simulation.
Money this week
The week's deals at a glance. The monthly review has the full picture.
Next week
What's coming from 25 October to 31 October, and how sure we are about each.
Stories are summarized in plain language; follow the link before quoting a figure. Story images are shown from the publishers' own sites.













