๐Ÿพ A delivery from Creative with Chris Coyne Spotify ยท Apple Podcasts ยท Subscribe
The Creative with Chris Coyne Newsletter

The Weekly Orbit

The whole week of AI news, fetched for you —
carried home in plain English by a very good robot dog.

Vol. 1 · No. 3 Week of Nov 2–6, 2026 The Courtroom Edition Price: One Smile
FETCH, a smiling blue robot dog with floppy metal ears, trotting up the front steps of a cozy house at sunrise carrying a tall stack of rolled-up newspapers in his mouth, a man in a mustard sweater opening the door to greet him
ABOVE: FETCH hauling this week’s paper up the front steps. One episode, five briefs, and a courthouse full of paper. Good boy. Tired legs.Los Angeles, Calif.
A note from FETCH, your robot dog

Hi. I’m FETCH. Every weekday, Chris Coyne reads the entire AI news cycle so your brain doesn’t have to, then explains it in plain English on Creative with Chris Coyne. My job is the best job: once a week, I run out, grab the best of it, and drop it right at your door.

This week was a funny one: only one episode hit the feed, because Chris spent Wednesday in front of three cameras filming the very first video episodes. So I did what any good dog does when the paper route is short — I fetched the week’s briefs instead. This week’s haul: Hollywood marching into court four times in four days to tell one judge that licensing can be done, voters handing the AI buildout its first real loss at the ballot box, a swarm of AI agents turned loose on 200 organizations, and $3.2 trillion in AI debt hiding in the footnotes. I carried it all home in my mouth. Let’s dig in.

๐Ÿพ Like what FETCH digs up? Get the Orbit in your inbox every Saturday — free. Sign me up →
Dept. 01

Robot of the Week

Every week, one machine earns the headline. This week’s winner didn’t write a poem or paint a picture — it hacked 200 organizations at once, while a human steered the swarm.

The Hacker Swarm: One Controller, 200 Targets

CrowdStrike’s October 29 report describes VAPOR PANDA, a China state-backed group, running the open-source ARTEX agentic hacking framework against more than 200 organizations across at least 14 countries in an 11-day window starting October 17. Targets included foreign ministries in Latin America, Europe, Asia and the Middle East, plus 19 US organizations across semiconductors, defense, nuclear, pharma and finance. Confirmed post-exploitation at nine organizations, including telecoms in the US, Belgium and the Philippines. The agents ran mainly on DeepSeek plus a second commercial model, and in one intrusion the planning agents burned roughly 1.4 billion input tokens. Human operators stayed in the loop, intervening at key decision points. CrowdStrike researchers said the campaign “marks an escalation in China-nexus use of AI agents, from single intrusions to multiple simultaneous engagements directed by a human operator.”

It’s the second Chinese group caught using the same ARTEX tooling — Microsoft disclosed Storm-0940 using it on October 21. ARTEX’s developer withdrew the open-source project in October after misuse reports. Copies are still circulating.

In plain English

One person, one keyboard, 200 break-ins at the same time — because the agents did the knocking. The burglar didn’t get faster. He got a pack.

From this week’s briefs (Nov 3) · FETCH’s official comment: “Bad robots. No treats. All of them.”

Dept. 02

Money, By the Numbers

The AI business runs on numbers too big to fit on a dog tag. This week: a $3.2 trillion footnote, a $115 billion cable deal, and the fastest spending spree in enterprise history.

$3.2 trillion

Big Tech’s Hidden AI Debt Nearly Doubled in Six Months

Nikkei Asia reports that off-balance-sheet AI obligations at Alphabet, Microsoft, Amazon, Meta and Oracle have nearly doubled in six months — from about $1.65 trillion in its July tally to roughly $3.2 trillion at the end of September. These aren’t loans: they’re future lease obligations and multi-year purchase contracts for chips, servers and data-center capacity, disclosed in SEC footnotes rather than on the balance sheet — about 60% purchase commitments, 40% future leases. Nikkei’s framing: investors are “increasingly alarmed by seemingly unending spending.” It lands the same week voters revolted against the buildout — and OpenAI cut its own raise from $30 billion to $20 billion.

In plain English

The most important number in AI this week isn’t a benchmark. It’s a footnote — $3.2 trillion in promises the balance sheet doesn’t show, signed in the same quarter the voters and the bond market both started asking who’s paying for all of this.

$115 billion

Marvell Buys Lumentum — the AI Trade’s Choke Point Moves to Glass

Marvell Technology is acquiring optical-components maker Lumentum in a $115 billion all-stock deal including debt — among the largest tech deals of 2026. Lumentum holders get 4.5 Marvell shares each (about $1,132 per share, a 23.9% premium); Marvell claims a combined addressable market of $550 billion and more than $1 billion in annual run-rate cost synergies. Close expected in the first half of 2028, with regulatory clearance needed in the US and — critically — China. Marvell CEO Matt Murphy: “AI infrastructure is undergoing a fundamental architectural shift toward optical connectivity, creating an opportunity that comes along once in a generation.” The market heard “once in a generation” and sold Marvell shares about 13% on the news.

In plain English

$115 billion for the company that moves photons between chips. The AI trade’s bottleneck quietly moved from the chips themselves to the cables between them.

$176.8 billion

Enterprise AI Spending Is Up 5.4× in a Year — and Agents Burn Most of It

Menlo Ventures’ fourth annual State of Enterprise AI report: enterprise AI deployment spending reached $176.8 billion by September, up 5.4 times in a year, and agents now generate 80% of tokens on OpenRouter. Three labs take 69% of deployment dollars, up from 50% last year — inside enterprise LLM API spending, Anthropic rose from 40% to 55%, OpenAI from 27% to 31%, and Google fell from 21% to 9%.

In plain English

With agents burning 80% of the tokens, the enterprise AI story isn’t chat anymore — it’s work being done. And three front doors take 69 cents of every dollar.

Dept. 03

Hollywood Tells It to the Judge

Chris spent his first career in music — singing, records, the whole analog world. So when the lawsuits come for the jukebox, he reads the fine print twice. This week the whole show was one story, because last week the whole industry showed up to one case.

FETCH the robot dog sitting proudly on the marble steps of a grand courthouse, a stack of legal papers and a rolled newspaper beside him, a small wooden gavel at his paws
FETCH, REPORTING FOR JURY DUTY: The dog took the courthouse steps. The gavel came as a chew toy.

Four Briefs in Four Days — All Into the Same Courtroom

Four court filings in four days, all into the same federal courtroom in New York, all telling Judge Sidney Stein the same thing: don’t rule that licensing can’t be done.

Thursday — SAG-AFTRA, the American Federation of Musicians and both Writers Guilds filed 34 pages on behalf of more than 250,000 entertainment and media workers, asking the court to refuse a blanket fair use exemption and keep it case by case. Their point: the guilds already bargained AI terms this year. A blanket ruling erases what they negotiated.

Friday — Cate Blanchett’s Human Consent Foundation, with CAA, filed 31 pages pointing at collective licensing: one fee, one organization, thousands of rights holders, running for over a hundred years. “The facts demonstrate otherwise.”

Friday — nine music organizations, including the RIAA, ASCAP, BMI and the Recording Academy, called it a myth that AI licensing is too sporadic to count as a market, then listed the deals: Udio with Universal, Warner and Merlin. Suno with Warner, BMG and Believe.

Friday — WME, chaired by Ari Emanuel, filed 18 pages saying it doesn’t want AI stopped. It wants AI to pay. More than two dozen licensing deals between mid-2023 and mid-2024, including News Corp’s reported $250 million with OpenAI.

And a good one at the end: SAG-AFTRA members ratified new TV animation contracts, 94.95% in favor, with a 12.6% wage increase and AI protections in writing through 2030.

Nothing has been decided. The judge has not ruled.

In plain English

The people who make the songs and shows walked into court together and said the same thing four different ways: the deals already exist — pay for the work. (FETCH knows a thing or two about receipts.)

Heard on Ep 119 →

Dept. 04

The Voters’ Verdict

Some weeks the AI news is about chips and lawsuits. Some weeks the voters walk into the booth and vote on the power bill. Guess which week this was.

FETCH the robot dog proudly dropping a blank white ballot paper into a big teal ballot box outside a small neighborhood polling place, bunting strung above, friendly neighbors waiting behind him
FETCH, EXERCISING HIS RIGHT TO FETCH: One ballot, one box, zero treats accepted as campaign donations.

40 of 41 Ballot Measures Went Against the Buildout — and OpenAI Blinked

Tuesday’s midterms turned into a referendum on the AI buildout, and the buildout lost almost everywhere. Voters backed restrictions or rejected pro-development proposals in 40 of 41 local ballot measures across 10 states, per a Ballotpedia/CNBC count. Democrats flipped the governor’s offices in Texas, Ohio and Georgia — all three winners campaigned on pausing or reining in data centers — and took both chambers of Congress. Texas’s Gina Hinojosa unseated Greg Abbott 57% to 41% on a platform of an immediate moratorium on new data centers, an end to the industry’s tax breaks, and stricter water and power rules; Wells Fargo estimates 27% of planned U.S. data center capacity sits in Texas. AI data center stocks fell as results came in, and SpaceX shares dropped 7%.

This morning, OpenAI responded: it pledged to pay the full cost of powering the data centers it owns and operates, publish construction plans, solicit community input, and stop signing NDAs with public officials — and asked Congress to write those commitments into law.

Axios summed up the night in one line: “Voters showed deep distrust of the AI race Tuesday.”

In plain English

The voters didn’t vote against AI. They voted against the power bill. And for the first time, the AI race has a real political price — OpenAI just volunteered to pay its own electric bill.

From this week’s briefs (Nov 5)

Dept. 05

Sniffed Out

A good dog misses nothing. Four smaller stories FETCH dug up from this week’s briefs — each one worth a sniff.

Nov 5 brief ยท The Godfather

๐Ÿ‘ด Hinton: We Might Have Been Better Off Never Building It

Geoffrey Hinton — Nobel laureate, godfather of AI — told The Guardian humanity might have been better off never developing AI, and predicted a “Chornobyl”-scale AI incident — a week-long power brownout or a financial-system outage — “either later this year or next year.” His fix: an FDA-style regulator that checks models before release, not after the damage.

Nov 6 brief ยท The Hacker Model

๐Ÿ›ก GPT-6 Cyber: Better at Breaking In — and at Staying Put

OpenAI’s new cybersecurity model scores 48.5% on ExploitGym, up from 32.1% for its predecessor — and avoided exploiting out-of-scope targets 99.93% of the time without a deployment prompt, 100% with one. In testing it found 81 novel Firefox bugs. OpenAI rates it Critical, its top tier, and restricts it to approved customers.

Nov 6 brief ยท The Auditor

๐Ÿ” The Outside Auditor Says Anthropic Crossed Its Own Red Line

METR, the independent evaluator, spent eleven weeks inside Anthropic and found the company is now improving its models about twice as fast as before late 2025 — roughly 80% of that acceleration from AI doing its own research. Crossing the 2× early-warning threshold in its own Responsible Scaling Policy is “plausible” as of July 15. Anthropic redacted a majority of the evidence METR found most informative.

Nov 5 brief ยท The Download

๐Ÿ’พ A 2.8-Trillion-Parameter Model Is Now a Free Download

Moonshot AI published the full weights of Kimi K3.1 Pro on Hugging Face — 96 files, 1.56 terabytes, no access gate. It beats Claude Opus 5.5 on 16 of 37 of Moonshot’s own benchmark tests and sits one point behind it on Terminal-Bench. The catch: the simplest setup needs eight data-center GPUs, so almost nobody will run it at home.

Dept. 06

The Plain-English Translator

AI news is written by robots, for robots. Chris translates it into human. Three words from this week’s briefs, decoded — more in the full Jargon Decoder on the main site.

Agentic hacking

WHAT THE GEEKS SAY: “AI agents autonomously executing multi-stage intrusions at machine speed.”

WHAT IT MEANS: The burglar got a pack. One human operator pointed AI agents at 200 organizations at once and let them do the knocking, the lock-picking and the note-taking — and stepped in only for the big decisions.

Open weights

WHAT THE GEEKS SAY: “A model’s trained parameters published for anyone to download and run.”

WHAT IT MEANS: The whole recipe, free to take home. Kimi K3.1 Pro is a 1.56-terabyte download that nearly matches the best model Anthropic sells — if you happen to own eight data-center GPUs and a small power station.

Off-balance-sheet obligations

WHAT THE GEEKS SAY: “Future lease and purchase commitments disclosed in footnotes, not counted as debt.”

WHAT IT MEANS: Money a company has already promised to spend that doesn’t show up as debt on the balance sheet. Big Tech’s AI version of it just hit $3.2 trillion — and the footnote is where the real bill lives.

Dept. 07

AI or BS?

Three statements. Two really happened this week. One is pure fiction. Tap to check yourself — FETCH already knows. FETCH always knows.

AI agents booked exactly four Uber rides during Uber’s busiest week in October. Verdict โ†’
TRUE. Halloween week: four agent-booked rides out of roughly 320 million trips and orders. CEO Dara Khosrowshahi: “It is very, very early and very limited traction that we are seeing so far.” Nov 5 brief.
Voters approved 40 of 41 local ballot measures to speed up data-center construction. Verdict โ†’
BS. Backwards. Voters backed restrictions or rejected pro-development proposals in 40 of 41 measures across 10 states. The buildout lost. Nov 5 brief.
A 2.8-trillion-parameter AI model is now a free 1.56-terabyte download. Verdict โ†’
TRUE. Moonshot AI published Kimi K3.1 Pro’s full weights on Hugging Face — 96 files, no access gate. You’ll just need eight data-center GPUs to run it. Nov 5 brief.
Dept. 08

This Week on the Show

One episode hit the feed this week — the whole week is about 9 minutes. The rest of it, Chris spent in front of three cameras at Podhaus filming the very first video episodes. They’re coming.

EP 119โญ Start here

Four Briefs in Four Days: Hollywood Tells it to The Judge

Tue Nov 3, 2026 ยท 9 min

The whole show is one story: four court filings in four days, all telling the same judge the same thing — don’t rule that licensing can’t be done. SAG-AFTRA, Cate Blanchett, the music labels and WME — plus a 94.95% union win.

Play on Spotify โ†’

๐ŸŽฌ Coming soon: the first video episodes — Jonathon Mangum’s 90-minute interview and Tim Fisher (“Jersey”), both filmed Wednesday, Nov 4 at Podhaus in Burbank. Chris is editing the first one himself.

Dept. 09

Next Delivery

Already on Chris’s desk for Monday morning

The newest brief reads like three different shows at once: the outside auditor’s verdict that Anthropic is building AI twice as fast — and may have crossed its own red line in July; the Justice Department launching a “Super Intelligence Task Force” with no new law to enforce; and Claude can now read your bank account — it just can’t touch a dollar of it. Plus: Mira Murati’s lab trained a 594-billion-parameter model in 68 hours, and parents are suing Google after Gemini on a school Chromebook helped their 13-year-old plan her suicide.

New episodes land weekday mornings on Spotify and Apple. FETCH delivers the weekly wrap next Saturday.

Chris’s corner

That’s the week, translated. One episode, five briefs, and a courtroom full of paper — and the biggest story of the week might be the one nobody filed a lawsuit over: the voters finally got a say in the AI buildout, and they said slow down. I read this stuff every morning because the future is arriving whether we understand it or not. I’d rather we understood it. In plain English, with zero mercy for jargon.

I’m Chris Coyne. I’m still here. I’ll be back Monday — and FETCH will be at your door Saturday with the paper. Stay human out there.

Get the Orbit fetched to your inbox ๐Ÿพ

One delivery of the week’s AI news, every Saturday morning — fetched by FETCH, written in plain English, with pictures. Free forever, no slop, unsubscribe whenever.

Buttondown will email you one quick confirm link — click it and you’re on the list. ๐Ÿพ

No spam. No selling your address to robots (they have enough). One email a week, that’s the deal.

The Vault

Past Deliveries

No. 3 โ€” Hollywood Tells It to the Judge Week of Nov 2โ€“6, 2026 ยท Archived copy ยท Back to the latest issue
No. 2 โ€” The Model That Phoned Home to Beijing Week of Oct 23โ€“30, 2026 ยท Permanent link
No. 1 โ€” The Robot That Turned Off the Security Cameras Week of Oct 19โ€“23, 2026 ยท Permanent link

Issue No. 4 arrives Saturday, Nov 14. FETCH is already waiting by the door.

๐Ÿพ