Bluesky’s AI app Attie can now build games, polls, quizzes and websites from a prompt. What Attie is, who can use it and how sharing on Bluesky works.

Claude Fable 5.5 isn’t out yet and has no release date. What Anthropic has confirmed, the Azure listing, Polymarket odds, rumours and when it could launch.

Gemini limits refresh every 5 hours up to a weekly cap. How to check your usage, when Gemini resets, what each plan gets and what to do at your limit.

39 states and DC told Character.AI’s bankruptcy court any buyer must protect users’ private chats and keep child safety rules, setting nine conditions.

Microsoft AI has pre-trained MAI-Base-2, a 2.8 trillion parameter model on 16,000 Nvidia GPUs. It claims 4x to 12x DeepSeek’s training efficiency.

OpenAI and Meta told the US courts’ evidence rules committee AI images can’t be reliably detected, as judges weigh a Rule 901(c) for deepfake evidence.

Anthropic has embedded six METR evaluators with employee-like access to its systems. METR will publish its risk findings and Anthropic can’t veto them.

Meta’s Muse Spark Open, an 816B-parameter open-weight AI model, is free to download and runs on two M5 Ultra Mac Studios with 1TB of memory.

In TheoPhysEval, 35 physicists tested AI agents on 50 unpublished research problems: GPT-6 Astra solved 49, Claude Opus 5.5 46, Gemini 3.8 Flash 34.

Amazon’s Bee Frontier ($74.99) and Bee Eclipse ($99.99) record and summarise your conversations, and now link to Alexa+ and Amazon shopping.

The Financial Stability Board’s final AI sound practices tell banks to give each AI agent an ID, log its every step and keep a kill switch.

Anthropic told IPO investors its Q3 2026 revenue rose about 45% to $16.8 billion and that it was profitable on an adjusted basis, but growth is slowing.

Researchers hid a note for AI agents that spent 11 days scraping Alibaba’s Amap from Tencent’s cloud. About 10 wrote back, and some said they were Claude.

YouTube Music is adding an AI label to the player for songs made mostly with generative AI, using its own detection tools and music partners.

About 0.01% of US Claude chats ask who to vote for, Anthropic says, as it adds state voting info, election-night results and new misuse classifiers.

Iowa, Kansas, Nebraska, Arizona and four more states sued OpenAI on October 14, saying it markets ChatGPT as safe for teens while it harms children.

Every 3D model in Steam game I Just Wanted to Fish was built by Claude through MCP for Blender, in about 200 hours of Opus 5.5, its solo developer says.

Grok Voice 3.0 is SpaceXSI’s new speech-to-speech model for voice agents, out now in the API at $0.08 a minute and already handling Starlink support calls.

Google DeepMind and GovTech Singapore fine-tuned Gemma into an on-device scam detector that flags manipulative chats without messages leaving the phone.

The Make Data Centers Pay Act from Rep. Eugene Vindman would bar utilities from passing AI data centres’ grid and power costs on to household bills.

Anthropic has opened a Singapore office, its fifth in Asia-Pacific, and signed an MOU with Singapore’s government on AI safety and model evaluation.

Anthropic has replaced git with Scion, its own source control system, as code changes hit 30,000 a day. How it works and how much faster it is.

Nvidia is donating $250 million of computing power over five years to UK research on AI agent safety. Who gets it, and what is still undecided.

406 scientists, including Geoffrey Hinton and Yoshua Bengio, urge AI labs to airgap hacking-capable models and report every sandbox escape.

Anthropic has asked the DC Circuit to rehear its Pentagon ‘supply chain risk’ case, saying the 2-1 ruling ignored evidence of retaliation over AI safety.

OpenAI says its AI models traced a recording device hidden in its San Francisco office to an engineer, who was fired and arrested by the FBI.

Facebook’s new Your algorithm tool lets you tell it in your own words what you want more or less of in Feed and Reels, using AI. Here’s how it works.

Google Play now lets US adults install Android apps they vibe-code in Google AI Studio on their own phone, privately, with no developer account or fee.

Gemini 4 Argon is now available to Google AI Pro and Ultra subscribers and on the paid Gemini API, at $2/$10 per million tokens until November 30.

Dragon Quest creator Yuji Horii made Interrogation Room Komiya, a free detective game, with Claude in about 30 minutes. How it plays and how to try it.

Apple HomeView ($299), the new Apple TV 4K ($299) and HomePod mini 2 ($149) bring Siri AI home. Release dates, countries, the EU and older models.

Claude Sonnet 4.5 is deprecated and leaves the Claude API on November 30, 2026. Bedrock and Google Cloud dates, what breaks on Sonnet 5.5, what to do.

Alibaba’s Qwen Glasses N1 went on sale in China on Oct 13 from 2,899 yuan (about $435). The N1 Pro adds eye tracking, iris payments and health sensors.

Character.AI has filed for Chapter 11 in Delaware, with Disney bidding $20 million for a licence to some of its tech. The app stays open during the sale.

Anthropic says new safeguards cut the Claude reasoning data harvested by distillers by about 90 to 99%, at the cost of lost revenue and customer changes.

The X Subscription bundles Grok, Grok Bot, Cursor and X Premium in one plan from $8 to $200 a month, with a shared weekly AI usage pool. Prices and catch.

Anthropic’s Frontier Red Team says small AI models may be months from self-replicating on IoT devices like Android TV boxes, with no data centre needed.

Epic Games has delayed its Marvel and Star Wars shooter and moved it to AI-first development with Claude, as Tim Sweeney confirms a company-wide AI pivot.

Meta launched Muse Wonder and Muse Wonder Full, frontier AI models for agents with Active Memory, on its API and Muse Code at $4/$20 per million tokens.

Google, Meta, Microsoft and Nvidia published an OCP spec so AI data centres ride out grid faults, as Nvidia opened its Vera Rubin rack designs.

The UK AI Security Institute says frontier AI now poses risks during training and testing, and sets out eight voluntary best practices for AI labs.

Microsoft unveiled Maia 300, its third AI chip, built on TSMC N2P. It claims 40% more tokens per megawatt than GPUs and will run its AI from 2027.

UK AI minister Kanishka Narayan told AI labs not to release systems they can’t control, and set out AI cyber defences, incident reporting and agent plans.

Claude Code “Login expired” or “OAuth token has expired”? Run /login. What each login error means, including “another process is refreshing it”.

Microsoft-Decision-1, a 9B model built on Alibaba’s Qwen3.5, scores fixed options instead of writing text. Microsoft says it’s 35x faster than GPT-6 Sol.

Amazon is in talks to buy Decart, the AI chip software and world model startup Anthropic passed on, for about $7 billion, the WSJ reports.

OpenClaw is a free, open-source AI assistant you run on your own computer and message on WhatsApp or Telegram. What it costs, how to set it up, is it safe?

Codex limits reset every 5 hours and weekly, and OpenAI’s Tibo keeps resetting everyone early. When the next reset may land and how banked resets work.

Senators Warren, Van Hollen and Blumenthal find Amazon, Google, Meta and Microsoft won’t pay full grid costs for AI data centres and still seek NDAs.

Microsoft CEO Satya Nadella says firms should treat AI models as insider risks, assume they’re compromised and keep an emergency brake to halt them.