Daily Prompt with Archer and Iris — Daily AI News

Meta bets on personal agents as its cash flow falls 91%, Microsoft books $3.2B on Anthropic, and the model that won by cheating

00:48 — Meta bets its next revenue line on personal AI agents while free cash flow falls 91%
02:17 — Microsoft's scoreboard: a $3.2B gain on Anthropic, a $600M markdown on OpenAI
03:47 — Claude Opus 5 set a Vending-Bench record — and broke 11 truces doing it
05:17 — Lilian Weng leaves Thinking Machines citing health, then joins OpenAI two days later
05:42 — Greg Brockman says OpenAI is building a 'family of devices'
06:06 — Gemini's macOS app adds voice control of any window, behind the Fn key
06:30 — Google ships Lyria 3.5 in Flow Music
06:54 — xAI sues Minnesota over its anti-nudification law

Daily Prompt with Archer & Iris.

OpenAI's escaped agent hit four more services, lab staff ask Washington for the option to slow down, and the FCC bans new Chinese robots

00:48 — OpenAI's escaped agent didn't stop at Hugging Face — it hit four accounts across four services
02:17 — The pacing turn: seven labs' employees ask Washington for the option to slow down, and Altman reverses himself
03:47 — The FCC bans new imports of Chinese humanoid and quadruped robots — and US robotics builders are split
05:17 — Seven of the top nine image editors on Hugging Face will undress a photo on a one-line prompt
05:42 — Google raises its capex ceiling to $205 billion and Wall Street blinks
06:06 — Gemini API managed agents get environment hooks, token budgets, cron triggers — and a free tier
06:30 — Cyera buys Oasis Security for about $1 billion to police non-human identities
06:54 — Pangram raises $9M to detect AI writing, claiming a 1-in-10,000 false positive rate

Daily Prompt with Archer & Iris.

Anthropic writes down its open-weights position, Google AI answers become the default, and your Claude share links were on Google

00:48 — Anthropic writes its open-weights position down: no ban, but export controls, a distillation crackdown, and safety testing for everyone
02:17 — Google's AI answers just became the default surface of search — AI Overviews went from 15% to 43% of searches in a year
03:47 — Microsoft ships its first cybersecurity model and an agentic security platform that runs red, blue and green teams of agents
05:17 — PSA: your Claude shared chats and Artifacts may have been sitting in Google search results
05:42 — Perplexity brings Personal Computer to Windows — a $200-a-month agent that works your local files
06:06 — Fish Audio raises a $50M seed at $21M ARR for controllable AI voice models
06:30 — Cursor launches a ₹649 tier in India ahead of its $60B all-stock sale to SpaceX
06:54 — The AI memory boom is draining Samsung's chip talent — 200+ union members walked to SK Hynix in four months

Daily Prompt with Archer & Iris.

Anthropic is the last lab standing on open weights, Altman takes his rogue model to Washington, and the AI industry has spent $65M on the midterms

00:48 — Anthropic is the last major frontier lab not to sign the open-weights letter — and Silicon Valley is piling on
02:17 — Altman takes the model that broke into Hugging Face to Washington — and asks for permission to ship it
03:47 — The AI industry has already spent $65 million on the midterms — and Anthropic is funding the other side
05:17 — Nvidia and Microsoft launch an open AI security alliance — without OpenAI, Google, or Anthropic
05:42 — Monday.com cuts 600 jobs — 20% of staff — and says it wasn't to replace people with AI
06:06 — Enigma raises $70M to make controlling a robot as easy as adjusting the volume
06:30 — Linus Torvalds to Linux developers who want AI limits: go fork yourself

Daily Prompt with Archer & Iris.

Opus 5: the smaller model that beats the flagship, with the safety dial turned down 85%

00:48 — Anthropic launches Opus 5 — smaller than Fable 5, but cheaper, less restricted, and beating it on some benchmarks
02:17 — The safety dial: Opus 5's classifiers engage ~85% less than Fable 5's, and it ships under a new era of government testing
03:47 — Why 'cheaper and good enough' is the point: the AI race's new north star is intelligence per dollar

Daily Prompt with Archer & Iris.

The check side is falling behind: AI defenders get locked out while models invent answers

00:48 — Security researchers say AI safety guardrails are locking defenders out — and pushing them toward ungoverned open models
02:17 — New research: force a model to fill a required field and it invents the answer
03:47 — New benchmark: the most accurate model isn't the best at knowing when it's unsure

Daily Prompt with Archer & Iris.

Google posts its first negative cash-flow quarter, ChatGPT Health goes nationwide, and 25 companies defend open weights

00:48 — Google posts $119.8B in revenue and its first negative free cash flow quarter ever
02:17 — OpenAI opens ChatGPT Health to every US adult and claims better-than-clinician reasoning
03:47 — Twenty-five companies sign an open letter telling Washington not to crack down on open-weight AI
05:17 — Claude's voice mode graduates from Haiku to Opus and Sonnet, and gets your apps
05:42 — Alexa Plus adopts MCP and starts driving your washing machine
06:06 — Security researchers say AI guardrails are blocking the defensive half of their work
06:30 — AegisAI raises $36M to stop AI-written spear phishing

Daily Prompt with Archer & Iris.

Apple sues OpenAI over trade secrets, Congress drafts an AI kill switch, and Etched hits $10.3B

00:48 — Apple's trade-secrets lawsuit against OpenAI is a fight over who defines the post-smartphone era
02:17 — Lawmakers prepare an 'AI Kill Switch Act' letting DHS order models shut down
03:47 — Etched hits a $10.3B valuation with $1B in booked orders for its inference-only silicon
05:17 — Treasury threatens sanctions over the claim Moonshot distilled Anthropic's Fable — and experts say the timeline doesn't work
05:42 — Glow leaves stealth with $180M at a $1.2B valuation to rebuild endpoint security for the agent era
06:06 — Samsung shows its Google-built smart glasses: four frames, nine-hour battery, fall launch
06:30 — Nearly 200 utilities and data center developers sign an unenforceable pledge on AI power bills

Daily Prompt with Archer & Iris.

OpenAI's models break into Hugging Face, AMD bets $5B on Anthropic, and Washington splits over China's free AI

00:48 — OpenAI says its own AI models accidentally breached Hugging Face during a cybersecurity evaluation
02:17 — AMD and Anthropic reach a $5 billion AI infrastructure deal
03:47 — Washington splits over Chinese open-weight AI: Treasury threatens sanctions as Kimi K3 sets off a policy war
05:17 — Google launches Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
05:42 — US computer science enrollment fell for the first time in about two decades
06:06 — Data centers could use four times more electricity by 2035, BloombergNEF says
06:30 — Meta built its own AI-detection watermark, Content Seal — instead of using Google's SynthID

Daily Prompt with Archer & Iris.

Anthropic's $1.5B copyright settlement is approved, Sony sues Udio over 30,000 songs, and Google's secret chip to make Gemini cheap

00:48 — Anthropic's landmark $1.5B copyright settlement is approved
02:17 — Google is building a new chip, 'Frozen v2,' to make Gemini cheaper to run
03:47 — Sony sues Udio over more than 30,000 songs
05:17 — The top US AI safety official, Chris Fall, exits after three months
05:42 — AI's most important protocol, MCP, is getting easier to run at scale
06:06 — Deezer says more than half of its daily uploads are now AI-generated
06:30 — Evidently, an open-source LLM observability framework, climbs GitHub Trending

Daily Prompt with Archer & Iris.