Hey crew, today two of our stories are about people struggling to describe AI, and one is about a joke that got the description exactly right.
No buzzwords, just what actually happened: OpenAI cut off Cursor because Musk bought it, the House Intelligence Committee wrote 56 pages on AI terrorism without a single concrete recommendation, and TikTok found the one format that makes people voluntarily share AI video. Practical reads, no hype, one clear takeaway per story.
Today's Stories:
The House Intelligence Committee's Latest 9/11 Report Has a Lot (and Nothing) to Say About AI
This Is the Kind of Video Your Mom Will See and Ask You if It's Real
OpenAI Ends Its Cursor Contract After SpaceX Buys It
Also in the mix: Nvidia's $3.5B into MediaTek · Malware hijacking Claude sessions · Instagram demoting AI influencers · Bankers on AI market manipulation
Let’s dive in.
WORLD
🕵️ The House Intelligence Committee's Latest 9/11 Report Has a Lot (and Nothing) to Say About AI
By Mike Pearl
© Alex Wong/Getty Images
Ahead of the 25th anniversary of the September 11 attacks, the House Permanent Select Committee on Intelligence, now led by Arizona Republican Rick Crawford, published a new review of the 9/11 Commission report. Roughly 10% of its 56 pages deal with AI, and the committee sounds worried.
The warnings center on AI-enabled terrorism. Models "have the possibility of making it significantly easier for any rogue actor, including terrorists, to develop and conduct more destructive attacks," the report says. Bioweapons get their own passage: knowledge that supports vaccine development and disease detection "may also be misused by hostile states or non-state actors."
The report also argues the intelligence community must build its own AI faster to stay ahead of adversaries already integrating it. It leans on testimony from Johns Hopkins professor Benjamin Buchanan, who invoked the "failure of imagination" that preceded 9/11, when scattered information across institutional seams produced a strategic surprise.
What the report does not include is a single specific measure. Its recommendation reads: policymakers "must carefully consider how to prevent AI tools from being leveraged for use by rogue actors," while the IC "must accelerate its own responsible adoption of advanced AI capabilities."
WORLD
📱 This Is the Kind of Video Your Mom Will See and Ask You if It's Real
By Matt Novak
A meme is spreading across TikTok and X built on a simple premise: post the most absurd AI video you can find, caption it with something like "Boomers gonna be praising Jesus over this one," and let the joke do the sharing. One clip shows a man riding an elephant that takes a ramp and performs feats no elephant can perform.
The joke lands on the videos as much as on the people who fall for them. AI clips rarely get shared on their own merit, and the mockery is doing what the content could not. A video of a "human statue" being kissed by a baby, posted in earnest by a Townhall Media editor, had real traction on X until a Community Note identified it as AI. The traffic collapsed immediately.
Elon Musk keeps posting AI clips of himself as a superhero and staring out windows. He told Joe Rogan last year that within five or six years, most of what people consume "will be just AI-generated content."
Detection is getting harder as the models improve. For now, viewers are voting with their attention, and the moment a clip is confirmed fake, they stop watching.
🌎 Elsewhere on the Internet
OpenAI cuts off Cursor after Musk buys it (link)
OpenAI is ending its contract with Cursor on November 12, four years into the partnership, because SpaceX now owns it. The stated reason is blunt: Musk's companies have broken OpenAI's terms before, and Musk admitted as much under oath. If you code in Cursor with GPT models, you have ten weeks to pick a new setup.Nvidia puts $3.5 billion into MediaTek (link)
The money comes as convertible bonds, and the partnership spans data centers, PC chips and cars. MediaTek will build custom AI accelerators that plug straight into Nvidia's racks. Another node in the web of cross-investments regulators keep calling circular financing.Anthropic says malware hijacked Claude login sessions (link)
A bad actor used malware to take over active Claude sessions, meaning the attacker walked in through an already-authenticated browser rather than the front door. Your AI account is now worth stealing, and your password alone won't protect it.Instagram will bury unlabeled AI influencers (link)
Synthetic accounts that don't disclose what they are will lose reach. For creators using AI faces or voices, the label is now cheaper than the algorithmic penalty. Disclose early rather than get demoted later.Bankers describe how AI could rig markets (link)
Central bankers laid out how autonomous trading agents could coordinate, herd, and move markets faster than any human could react. Pairs with today's Bank of England letter to the G20, and the two arrived within days of each other.
🛠️ Your AI Toolkit
Studio: browser video editing that feels like design.
Typed prompts: generate briefs from messy notes.
Link inbox: collects everything you mention across platforms.
💡 The Last Byte
A podcaster called two OpenAI agents Philip and Alexander this weekend, and the internet spent 48 hours arguing about whether bots can have a civilization.
📬 Prompt of the Day: Photograph the Fridge
You have chicken, half a lemon, and something in a container you're afraid of. Recipe sites want you to shop first.
"Here's a photo of my fridge. Give me three dinners I can make tonight using only what you can see, plus salt, oil and pepper. Rank them by how fast they are."
Why it works: you're describing the constraint instead of the goal, so nothing comes back requiring a trip to the store. Adding "rank by speed" matters at 7pm on a Tuesday, when the deciding factor isn't which meal is best but which one is done first.
🔢 The Big Number: $83,000,000
That's roughly what OpenAI's ads business is pulling in per month, the figure behind today's headline $1 billion annualised run rate. The number worth watching is the gap: to hit its stated $2.5 billion target for 2026, OpenAI needs to average over $541 million a month for the rest of the year. That's a 6.5x acceleration in four months.
⚙️ Decoder: “Frontier Model”
A frontier model is simply one of the handful of most capable AI systems in existence at any given moment, the ones pushing the edge of what's possible. GPT-5, Claude, Gemini, Mythos.
See you tomorrow.
Gizmodo