THE RUNDOWN

Newly unsealed court documents from The New York Times' copyright case against OpenAI and Microsoft reveal something uncomfortable — the companies' own internal documentation warned they were creating what they called a 'doom loop' for the open web. Essentially, AI systems scrape web content to train, which degrades the quality of that content over time, which then degrades the AI, which needs more scraping. They knew. If you've been following the AI training data debate, this one lands hard.

Governor Gavin Newsom is pushing legislation that would require frontier AI models to have a mandatory 'kill switch' — a way to shut them down fast if something goes wrong. California is positioning itself as the de facto national AI regulator, and this move puts real teeth on the table. Given the military hallucination story above, the timing feels almost deliberate.

Three independent researchers at Hacktron say it took them less than 72 hours to breach OpenAI employee accounts — and they did it using Anthropic's Claude as part of the attack chain. It's a wild irony: one AI company's tool used to compromise another's. The story is a sharp reminder that AI-assisted hacking isn't a future threat; it's the present one.

Anthropic has quietly set up a lab that conducts actual biology experiments, not just computational ones. This is the same company whose researchers have published warnings that AI could pose existential risks to humanity. The tension between 'AI will cure disease' and 'AI might kill us all' is apparently happening inside the same building now.

Disney — yes, that Disney, a company that's been around for over a century — has just hired its very first Chief Technology Officer, and the pick is the former CEO of Character.AI, the AI chatbot company Disney itself previously accused of copyright infringement. The headlines write themselves. It signals how seriously the entertainment giant is taking AI, even if the optics are a bit chaotic.

The FCC has waived its foreign ownership cap — normally capped at 25% equity — to allow sovereign wealth funds from Saudi Arabia, the UAE, and Qatar to collectively hold a 49.5% stake in the Paramount-Warner Bros. merger entity. That means three Gulf state governments will have significant influence over CBS and some of the biggest media brands in the world. The FCC's stated concern this week, meanwhile, was reportedly about who The View invites as guests — not this.

The big camera story on this year's iPhone 18 Pro is a variable aperture main lens — the kind of hardware flexibility that lets you physically control how much light enters the sensor, not just do it in software. It's the sort of feature that photographers have wanted on a smartphone for years, and it puts Apple in direct conversation with dedicated camera makers in a way previous iPhones haven't quite managed.

Separate from the military story, the FAA has announced plans for an $875 million AI tool to help manage air traffic congestion, starting with the notoriously complex DC airspace before a nationwide rollout. Given the day's earlier headlines, the optics of rushing AI into high-stakes infrastructure decisions feel complicated — but the congestion problem is real and the FAA says the tool has been carefully tested.

India's government is now requiring apps like Truecaller to pipe their spam reports directly to telecom operators. Truecaller isn't happy — it's essentially being asked to hand over commercially valuable proprietary data it spent years building, for free, to companies that are also its competitors in some markets. It's a pattern worth watching across Africa too, as regulators look at how much data private tech companies should be required to share with state-linked telecoms.

It's a heavy Friday in tech news. Between AI nearly starting wars, hallucinations in the courtroom, and Gulf petrodollars buying into Western media, the week ends with more open questions than answers. That's usually when the most interesting conversations start.