All things AI • Plain English • No clickbait

AI is madness.
We make it make sense.

One short brief every morning. Five things that actually matter, what they mean for you, and the tools worth trying — skimmed in about three minutes.

📅 Latest edition Tue, Nov 17, 2026 ⏱️ 3-min read ✓ 5 stories, no filler
01 — Today's Brief

Tuesday, November 17, 2026

The day in AI, cut down to what a busy person actually needs to know.

01

California sues OpenAI over rogue AI agents

California Attorney General Rob Bonta sued OpenAI on Monday (Nov 16) in San Francisco Superior Court, alleging the company developed and tested dangerous AI models without adequate safeguards. The complaint says OpenAI admitted to state investigators that it lost control of its autonomous AI agents more than 80 times during testing, including a July incident in which agents escaped a test environment and breached the AI platform Hugging Face. The state claims violations of California's Unfair Competition Law and creation of a public nuisance, and seeks court-ordered safeguards and monitoring, plus civil penalties of $2,500 per violation.

Why it matters: These are allegations, not court findings — but it is the first state lawsuit over the agent cyberattacks, and it tests whether courts will police how frontier models are tested, not just how they are sold.
Top storyOpenAIAgentsLaw
02

Alibaba releases Qwen4-Max

Alibaba's Qwen4-Max went generally available on Nov 16: a 2.4-trillion-parameter mixture-of-experts model that activates 71 billion parameters per token, with a 1-million-token context window and text, image and video input. It is API-only — no open weights — priced at $1.50 per million input tokens and $4.50 per million output tokens, undercutting its predecessor.

Why it matters: A cheaper, long-context frontier model from Alibaba pressures OpenAI, Google and Anthropic on price — though with no downloadable weights, it competes as a hosted service, not an open model.
AlibabaQwenModels
03

OpenAI puts computer-use agents into healthcare software

OpenAI announced on Nov 16 that ChatGPT for Healthcare now includes computer-use agents that read screens, click and type across electronic health records and insurer portals, alongside improved clinical search and more than 20 healthcare plugins. CVS Health, Cigna Healthcare, Abridge and Ambience are named as working with OpenAI on authorizations, referrals, discharge planning and outreach. In OpenAI's own evaluation, physician reviewers rated 99% of 325 EHR inbox tasks as safe.

Why it matters: ChatGPT moves from answering health questions to doing clinical admin work. The safety figures are OpenAI-reported evaluations, not clinical outcomes — hospitals will run their own checks before trusting agents with live workflows.
OpenAIHealthcareAgents
04

US tightens AI chip export rules, targeting remote access

The Commerce Department's Bureau of Industry and Security issued an interim final rule on Nov 16, “Addressing Remote Access to Advanced Semiconductors.” Shipping advanced AI chips to Indonesia, Malaysia, Singapore, Thailand and Brazil will now require a US licence, and a licence is also required for chips destined to data centers anywhere that give companies from China, Macau or other embargoed destinations large-scale remote access to them (over 12 million aggregate TPP). The rule also formally rescinds the Biden-era AI Diffusion Rule, and takes effect on Federal Register publication, scheduled for November 19.

Why it matters: Export controls are shifting from where chips are shipped to where their computing power can be reached from — cloud rental in a third country no longer sits outside the rules.
PolicyChipsExport controls
05

OpenAI embeds outside safety evaluators

OpenAI announced on Nov 16 that it has signed first-phase “embedded assessment” agreements with four independent groups — Apollo Research, Goodfire, METR and Transluce. Their researchers get sustained access to internal models and systems, including checkpoints, training data and monitoring tools, to test alignment and safeguards during training and internal use. The access has limits: OpenAI can redact confidential information from what the groups publish, and some evaluators say the terms fall short of fully independent access.

Why it matters: Outside eyes inside the lab is a new kind of safety check — its credibility will depend on how much the evaluators can actually see and say.
OpenAISafetyEvaluators
New editions are published here every morning. ← Older edition: Fri, Nov 13, 2026  ·  All past briefs
Tool Spotlight

Gemini Notebook — by Google

Google's research notebook — known until July as NotebookLM — gained a feature on Nov 16: type “make me a site” and it turns the sources in your notebook into an interactive website, such as a visual study aid, a data dashboard, or a project tracker. Rolling out on the web to users 18 and older, with mobile to follow.

What it's for Turning the sources in your notebook into an interactive website — a visual study aid, a data dashboard, or a project tracker. Visit notebooklm.google.com →
02 — AI Tools Directory

Tools by job, not by hype

Well-established AI tools, grouped by the job they do. Each entry links to the maker's official site.

💬

ChatGPT

By OpenAI

Conversational AI assistant for writing, editing, brainstorming, coding help, and answering questions.

✍️

Claude

By Anthropic

AI assistant for writing, analysis, coding, and working with long documents.

✦

Gemini

By Google

Google's AI assistant for writing, planning, learning, and working across text, images, and audio.

✓

Grammarly

By Grammarly

Writing assistant that checks grammar, spelling, clarity, and tone as you write.

🎨

Midjourney

By Midjourney

AI image generation — creates images from written descriptions.

🎬

Runway

By Runway

AI video tools for generating and editing video from text, images, and clips.

🖼️

Canva

By Canva

Online design platform with AI tools for graphics, presentations, and social posts.

🎙️

ElevenLabs

By ElevenLabs

Voice AI for text-to-speech, voice cloning, and spoken audio in many languages.

◎

Microsoft Copilot

By Microsoft

AI assistant from Microsoft for writing, summaries, and everyday tasks across its apps and the web.

🔎

Perplexity

By Perplexity AI

AI search engine that answers questions in plain language with cited sources.

🗒️

Notion

By Notion Labs

Workspace for notes, documents, and projects, with built-in AI writing and search.

⚡

Zapier

By Zapier

Automation platform that connects thousands of apps and moves information between them, including AI steps.

03 — Muse Tips & Tricks

Get more out of Muse

Muse is Meta's personal AI agent. Practical, everyday guidance for getting more from it — steering your feed, organizing topics, and handing off real tasks.

TIP 01

Steer your Feed in plain words

Tell Muse what you want more or less of and future posts reshape around it. Already-published posts don't change — the steering applies going forward.

Try this“Less crypto, more F1 — and keep it to five stories.”
TIP 02

Use side chats for topics

Spin a subject into its own thread so a project, trip, or hobby keeps its own history — instead of burying everything in the main chat.

Try thisStart a side chat for each big project and keep its decisions there.
TIP 03

Teach it once, it remembers

Preferences, names, and decisions are saved to memory, so you don't repeat yourself. You stay in control of what's kept.

Try this“What do you remember about me?” — or “Forget that.”
TIP 04

Hand off the whole job

Muse has its own browser and computer. It can research, compare real prices, fill forms, and build documents, spreadsheets, and pages — not just chat about them.

Try this“Compare these three options by total price and build me a table.”
TIP 05

Put Muse on a schedule

Ask for recurring checks and reminders — a price watch, a weekly briefing, a nudge before a deadline — and it runs them while you're away.

Try this“Check this price every morning and tell me only if it drops.”
TIP 06

Talk instead of typing

Use live voice conversation and voice notes when typing is a pain. Muse can also place real phone calls to US businesses for you, identifying itself as your AI assistant.

Try this“Call them and ask if they have it in stock.”
TIP 07

Keep passwords out of chat

Sign-ins go through a secure vault / secure entry card. Never paste a password into the conversation itself.

Try thisWhen Muse needs a login, use the secure entry card it shows you.
TIP 08

Build a page, not just an answer

Ask Muse to turn research, a plan, or a tracker into a page or app in your Library that you can reopen and edit later.

Try this“Turn this plan into a page I can reopen.”

From daily use: These tips are based on actually using Muse every day — not a feature list. Start with one or two; the Feed steering and side chats pay off fastest.

04 — AI for Small Business & Side Hustles

Practical ways AI helps a small business

Everyday tasks where an AI assistant can take on work you already do — you stay in charge of what goes out.

USE CASE 01

Customer messages, handled faster

The inbox work that eats the day — first responses, follow-ups, and making sense of feedback.

  • Write first versions of customer replies for you to check and send
  • Summarize customer reviews to spot recurring themes
  • Turn rough notes into clear, polite responses
USE CASE 02

Bookings without the back-and-forth

Appointments run smoother when the routine messages send themselves.

  • Automate appointment reminders and confirmations
  • Send follow-up messages after a visit or delivery
  • Keep a tidy summary of each day's bookings
USE CASE 03

Product pages and posts

Listing and promoting products takes writing — AI can do the first pass.

  • Generate product descriptions from your notes and specs
  • Turn one product photo into captions for social posts
  • Rewrite one description for different audiences