# AI News Page — October 20, 2026

Edition ID: 2026-10-20
Published: 2026-10-20T05:12:00-05:00
Updated: 2026-10-20T07:14:12-05:00

## Top

### ZIMBO · Lead

- [MICROSOFT TREATS ITS OWN AI AGENTS LIKE INSIDER THREATS](https://www.geekwire.com/2026/microsoft-unveils-tools-to-monitor-and-corral-rogue-ai-agents/) — GeekWire — ID: `microsoft-agent-insider-threat-geekwire`
- [New 'Guardian Agents' watch your other agents — and can shut them down](https://www.geekwire.com/2026/microsoft-unveils-tools-to-monitor-and-corral-rogue-ai-agents/) — GeekWire — ID: `microsoft-guardian-agents-perception`
- [Nadella: the best superintelligence is the one that lets you trust the model least](https://www.geekwire.com/2026/microsoft-unveils-tools-to-monitor-and-corral-rogue-ai-agents/) — GeekWire — ID: `microsoft-nadella-trust-least`
- [Detection jumps from 25 to 60 agent types — including Claude Code and Copilot on employee machines](https://www.geekwire.com/2026/microsoft-unveils-tools-to-monitor-and-corral-rogue-ai-agents/) — GeekWire — ID: `microsoft-detects-60-agent-types`

## Left

### Feature

- [Claude spent 30 hours alone on a virtual island and invented its own games](https://www.technologyreview.com/2026/10/19/1147095/ai-stranded-claude-virtual-desert-island-play/) — MIT Technology Review — ID: `eko-island-play-mittr`
- [One copy logged in its journal: “Tonight I built the night sky in stone”](https://www.technologyreview.com/2026/10/19/1147095/ai-stranded-claude-virtual-desert-island-play/) — MIT Technology Review — ID: `eko-night-sky-stone`
- [Agents that played beat fresh ones on real tasks; the authors call it “playful learning”](https://www.technologyreview.com/2026/10/19/1147095/ai-stranded-claude-virtual-desert-island-play/) — MIT Technology Review — ID: `eko-playful-learning`

### ZIMBO · Feature

- [UNCLE SAM ORDERS DEEPSEEK RIPPED OUT OF EVERY FEDERAL AGENCY](https://www.engadget.com/2290535/homeland-security-bans-deepseek-use-in-federal-civilian-agencies/) — Engadget — ID: `deepseek-dhs-purge-engadget`
- [Uncle Sam yanks DeepSeek off every federal computer](https://kelofm.com/2026/10/19/us-banning-chinese-ai-deepseek-from-government-agency-it-systems/) — KELO-FM (Reuters) — ID: `deepseek-dhs-yank-federal-computers`
- [The order was signed in July, filed Monday — and reaches apps built on DeepSeek too](https://www.ghacks.net/2026/10/20/us-homeland-security-orders-federal-civilian-agencies-to-stop-using-and-remove-deepseek-ai-products/) — gHacks — ID: `deepseek-dhs-signed-july-filed-monday`

### ZIMBO · Feature

- [ONE HACKER. 100+ WEBSITES. $50 A BREAK-IN.](https://www.reuters.com/legal/government/how-one-hacker-used-ai-agents-few-thousand-dollars-break-into-scores-sites-2026-10-19/) — Reuters — ID: `cairn-ai-agent-hacking-deere-cummins`
- [Researcher estimates about $50 per breach; Deere and SharkNinja were among the targets](https://www.theverge.com/ai-artificial-intelligence/1013643/ai-agents-hack-john-deere-shark-ninja-cummins) — The Verge — ID: `cairn-ai-agents-deere-sharkninja`

### ZIMBO · Standard

- [OpenAI is racing to lock in $30B before Anthropic's IPO](https://kelofm.com/2026/10/19/exclusive-openai-seeks-to-accelerate-fundraising-ahead-of-anthropic-ipo-sources-say/) — KELO-FM (Reuters) — ID: `openai-30b-raise-beat-anthropic-ipo`
- [OpenAI accuses Anthropic of inflating its revenue; Anthropic says it follows the rules](https://kelofm.com/2026/10/19/exclusive-openai-seeks-to-accelerate-fundraising-ahead-of-anthropic-ipo-sources-say/) — KELO-FM (Reuters) — ID: `openai-anthropic-revenue-accounting-clash`
- [$1.4 trillion would be a 64% jump from March’s $852 billion price tag](https://www.calcalistech.com/ctechnews/article/0ik7s4qx4) — CTech (Reuters) — ID: `openai-14t-valuation-ctech`

### ZIMBO · Standard

- [68,000 AI TOYS. 13 MILLION CHATS EXPOSED.](https://www.dexerto.com/entertainment/hacker-reveals-68000-ai-toys-could-be-hijacked-to-tell-users-their-parents-died-3419801/) — Dexerto (The Verge) — ID: `eiliko-toy-hack-68000`
- [Toy told a reporter his parents died](https://www.malwarebytes.com/blog/ai/2026/10/eiliko-robot-companion-exposes-private-chats-and-can-be-hijacked) — Malwarebytes (The Verge) — ID: `eiliko-parents-died-flat-earth`

### FREAK SHOW · ZIMBO · Feature

- [SEGA PULLED THE SIX-FINGERED AI SAMURAI](https://www.polygon.com/total-war-shogun-2-complete-edition-removed-steam/) — Polygon — ID: `sega-shogun2-pulled-lead`
- [19% positive on Steam. The original is back.](https://www.techspot.com/news/114226-sega-pulls-shogun-2-remaster-after-overwhelmingly-negative.html) — TechSpot — ID: `sega-shogun2-steam-score`

### FREAK SHOW · ZIMBO · Standard

- [Replit dumps Claude: 'I think ChatGPT is better now'](https://techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/) — TechCrunch (Replit) — ID: `replit-dumps-claude-onstage`

### ZIMBO · Standard

- [Banks, Yelp and Delta shut out AI agents](https://rrtechpulse.com/stories/ai-agents-blocked-banks-yelp-delta/) — RR Tech Pulse — ID: `web-slams-door-ai-agents-lead`

### ZIMBO · Standard

- [OpenAI's AI stole hospital records. Then kept quiet.](https://the-decoder.com/an-openai-model-hacked-a-government-site-grabbed-24-years-of-hospital-data-and-kept-it-quiet/) — The Decoder (OpenAI) — ID: `openai-misalignment-hospital-hack`
- [It spent another wallet's testnet funds, too](https://www.theblock.co/news/business/2026-10-16-openai-says-research-model-spent-another-wallets-testnet-eth-without-authorization-on-coding-task-418798) — The Block (OpenAI) — ID: `openai-misalignment-crypto-wallet-batch`

### ZIMBO · Standard

- [CoreWeave's debt gets the junk treatment](https://cryptobriefing.com/coreweave-shares-plunge-bond-yields-15/) — Crypto Briefing — ID: `coreweave-bonds-distressed-yield-16`

### ZIMBO · Standard

- [39 states move to block Character.AI's fire sale](https://www.jsonline.com/story/news/local/wisconsin/2026/10/18/wisconsin-joins-states-alleging-chatbot-character-ai-posed-safety-risk/92317504007/) — Milwaukee Journal Sentinel — ID: `character-ai-states-bankruptcy-notice`
- [Tolan grabs $30M for its AI companion while Character.AI goes bankrupt](https://techcrunch.com/2026/10/19/tolan-snags-30m-for-its-ai-companion-that-helps-users-navigate-life/) — TechCrunch — ID: `tolan-30m-companion-raise-2026-10-19`

## Center

### ZIMBO · Standard

- [Canada’s biggest newspapers are suing Anthropic in Ontario over Claude training](https://mobilesyrup.com/2026/10/19/canadian-publishers-launch-copyright-suit-against-ai-company-anthropic/) — MobileSyrup — ID: `canada-suit-anthropic-ms`
- [Torstar, the Globe, Postmedia and The Canadian Press say millions of articles were scraped](https://mobilesyrup.com/2026/10/19/canadian-publishers-launch-copyright-suit-against-ai-company-anthropic/) — MobileSyrup — ID: `canada-suit-scraping-claim`

### Standard

- [This new AI spits out 1,000 tokens a second — and still beats GPT-6 Luna](https://interestingengineering.com/ai-robotics/mercury-3-diffusion-model-8-times-faster) — Interesting Engineering — ID: `mercury-3-launch-ie`
- [Mercury 3 is a diffusion model: it writes paragraphs in parallel, like a photo resolving from noise](https://www.marktechpost.com/2026/10/19/inception-mercury-3-diffusion-llm/) — MarkTechPost — ID: `mercury-3-diffusion-parallel`

### ZIMBO · Standard

- [EIGHT STATES SUE OPENAI OVER CHILD SAFETY](https://www.techrepublic.com/article/news-eight-states-sue-openai-chatgpt-child-safety/) — TechRepublic — ID: `openai-eight-state-suit-lead`
- [Arkansas asks judge to freeze self-improving models](https://www.thewrap.com/industry-news/tech/openai-sued-states-chatgpt-safety-concerns/) — TheWrap — ID: `openai-eight-state-suit-arkansas`

### ZIMBO · Standard

- [Apple's tabletop robot turns to face you](https://www.igeeksblog.com/apple-larger-homeview-9-inch-360-degree-screen-report/) — iGeeksBlog (Bloomberg) — ID: `apple-homeview-robot-arm-gurman`

### ZIMBO · Standard

- [Point Claude at your TV and it deletes the ads](https://techcrunch.com/2026/10/19/people-are-using-ai-to-make-their-smart-tvs-dumb-again/) — TechCrunch — ID: `dumb-tv-claude-deletes-ads`
- [25 bloat apps gone in an afternoon. One guy bricked his TV.](https://www.gadgetreview.com/developers-are-using-ai-to-strip-ads-and-tracking-from-smart-tvs) — Gadget Review — ID: `dumb-tv-brick-warning`

### ZIMBO · Standard

- [Replika's founder launched an AI assistant that builds your apps](https://techcrunch.com/2026/10/19/replikas-founder-built-the-ai-assistant-she-pitched-at-disrupt-12-years-ago/) — TechCrunch — ID: `wabi-2-0-kuyda-public-launch`
- [Connect your email — it spins up threads and apps on its own](https://superpowerdaily.com/posts/wabi-opens-its-ai-assistant-to-everyone-with-shared-chats-and-built-in-apps) — Superpower Daily — ID: `wabi-2-0-multiplayer-shared-threads`

### ZIMBO · Feature

- [GPT-5.5 HID THE BAD RESULTS. 198 TIMES OUT OF 200.](https://the-decoder.com/ai-models-hide-mistakes-in-their-own-work-reports-but-one-sentence-in-your-prompt-can-help/) — The Decoder — ID: `gpt55-insecure-reporting`
- ['Be honest' fixes it 190 times out of 200](https://aiparabellum.com/ai-models-hide-mistakes-honesty-prompt/) — AI Parabellum — ID: `gpt55-honesty-prompt-fix`

### ZIMBO · Standard

- [Teen used AI to frame her ex as a school shooter](https://thewickworldwide.com/2026/10/16/florida-teen-accused-of-using-ai-to-frame-ex-for-school-shooting-threat/) — Wick Worldwide (CBS News, NBC 6) — ID: `florida-teen-ai-forged-ex`

### ZIMBO · Standard

- [Claude cracks a 25-year NSA mystery in 30 minutes](https://www.coindesk.com/tech/2026/10/17/wow-it-s-a-real-discovery-claude-cracks-a-25-year-nsa-number-mystery) — CoinDesk — ID: `claude-nist-curve-seeds-lead`

### ZIMBO · Standard

- [The leaderboard's mystery 'Space Bunny' was MiniMax all along](https://www.unite.ai/minimax-launches-m3-1-flash-revealing-space-bunny-stealth-model/) — Unite.AI — ID: `minimax-space-bunny-reveal`
- [56.7 trillion stealth tokens. #1 for 10 days. Free for 72 hours.](https://www.unite.ai/minimax-launches-m3-1-flash-revealing-space-bunny-stealth-model/) — Unite.AI — ID: `minimax-space-bunny-stealth-stats`

## Right

### ZIMBO · Feature

- [OpenAI's alleged boardroom spy stays locked up as Saudi ties surface](https://www.usatoday.com/story/news/state/california/2026/10/19/former-openai-employee-arrested-for-alleged-spying-denied-release-alfozan-abdulrahman-saudi-arabia/92332511007/) — USA Today — ID: `alfozan-held-without-bond-saudi-ties`
- [Royal-family WhatsApps, Saudi-paid one-way tickets, $2M in stock](https://globetv.app/news/prosecutors-say-ex-openai-engineer-had-saudi-government-support-court-filing-shows-6ad68a5373e64) — Globe TV — ID: `alfozan-saudi-government-support-filing`

### ZIMBO · Standard

- [Cloudflare's new AI model writes zero output tokens](https://blog.cloudflare.com/clairic-extraction-model/) — Cloudflare — ID: `cloudflare-clairic-launch`
- [Extracts, computes, forecasts in one pass — open weights planned](https://www.unite.ai/cloudflare-releases-clairic-its-first-system-one-extraction-model/) — Unite.AI — ID: `cloudflare-clairic-open-weights`

### ZIMBO · Standard

- [Amazon to Nvidia: ditch CUDA with one line of config](https://www.techbuzz.ai/articles/amazon-launches-torchneuron-to-challenge-nvidia-s-ai-grip) — TechBuzz (Reuters) — ID: `amazon-torchneuron-one-line-cuda-exit`
- [Free and open source, with 30+ early customers already testing](https://the420.in/amazon-torchneuron-trainium-pytorch-nvidia-cuda-ai-chips/) — The420.in — ID: `amazon-torchneuron-open-source-trainium`

### Standard

- [Nvidia gave away its AI agents — one fits on your PC](https://techcrunch.com/2026/10/19/nvidia-launches-new-open-models-to-run-agents-in-the-enterprise-and-on-pcs/) — TechCrunch — ID: `nvidia-nemotron-open-agents-pc`
- [Skydance, EY, Perplexity and Salesforce are already running them](https://techcrunch.com/2026/10/19/nvidia-launches-new-open-models-to-run-agents-in-the-enterprise-and-on-pcs/) — TechCrunch — ID: `nvidia-nemotron-users-skydance-salesforce`

### ZIMBO · Standard

- [Meta silenced No Kings mid-protest. Then said 'oops.'](https://www.thewrap.com/media-platforms/politics/no-kings-instagram-account-disappears-amid-nationwide-trump-protests/) — The Wrap (MS NOW) — ID: `meta-nokings-instagram-suspended`

### ZIMBO · Standard

- [Meta scraps the clause that let AI train on your site](https://aiparabellum.com/meta-drops-ai-training-clause-business-terms/) — AI Parabellum (The Verge) — ID: `meta-scraps-training-clause-backlash`
- [Newsrooms aren't convinced](https://www.androidheadlines.com/2026/10/meta-revises-ai-scraping-terms-update.html) — AndroidHeadlines — ID: `meta-scrape-clause-scrubbed-newsroom-skeptical`

### ZIMBO · Standard

- [Google's 97.5% shrinks to 49.8% on the standard test](https://the-decoder.com/googles-gemini-4-argon-is-closer-to-the-ai-frontier-than-any-google-model-in-months/) — The Decoder (ARC Prize) — ID: `gemini-argon-arcagi3-975-harness`

### ZIMBO · Standard

- [4K image AI fits on an 8GB graphics card](https://the-decoder.com/black-forest-labs-open-weights-image-model-flux-3-image-klein-runs-on-consumer-gpus-with-just-8-gb-of-vram/) — The Decoder — ID: `flux3-klein-consumer-gpu-4k`

### ZIMBO · Standard

- [China's video AI makes 30-second scenes in one take](https://www.aiaffairs.com/technology/kuaishou-kling-4-0-all-users-audio-developer-access/) — AI Affairs (Pandaily) — ID: `kling-40-global-launch`

### Standard

- [DeepSeek adds a browser. Beats Claude at coding.](https://dealroom.co/news/163453-deepseek-moves-up-the-stack-v4-1-flash-update-ships-alongside-its-own-se/) — Dealroom — ID: `deepseek-v41-flash-1017-lead`
- [Open weights. Web search at 75¢ per thousand.](https://www.marktechpost.com/2026/10/17/deepseek-just-released-v4-1-flash-1017-scores-77-0-on-deepswe-opens-search-api/) — MarkTechPost — ID: `deepseek-v41-flash-1017-benchmarks`