The Sam Altman Biopic Just Got Fired
Amazon MGM has dropped Artificial, Luca Guadagnino’s film about Sam Altman’s 2023 firing-and-reinstatement saga, which starred Andrew Garfield in the Altman role. The irony of Amazon dropping a project about the drama between Altman and his board is almost too on-the-nose given Amazon’s massive investment in OpenAI’s competitor Anthropic — though I’m sure that’s a coincidence. Either way, given that OpenAI is now barreling toward an IPO and Altman’s grip on the company has never been tighter, a film that frames him as the embattled underdog might have felt increasingly awkward to finance.
Anthropic’s Government Ban Is Accidentally a Marketing Campaign
When the US government forced Anthropic to pull Fable 5 and Mythos 5 from release citing national security concerns, the expectation was presumably that this would hurt Anthropic. Instead, cybersecurity researchers signed an open letter calling the move dangerous, and Anthropic’s usage numbers apparently didn’t notice the ban at all. Nothing says “your AI is powerful and trustworthy” quite like the federal government panicking about it — the Streisand Effect has entered the enterprise AI market.
A History Lesson on Why AI Export Controls Won’t Work Either
TechCrunch traces the through-line from PGP encryption export battles in the ’90s to today’s attempt to contain Anthropic’s Mythos cybersecurity model, and the conclusion is pretty damning: the US government has a thirty-year losing streak at keeping software bottled up. The question isn’t really whether these controls work — they demonstrably don’t — but why policymakers keep reaching for the same broken tool. Spoiler: it’s because doing something feels better than admitting you can’t control the thing.
A Startup Claims It Solved a Decade-Old LLM Bottleneck
Miami-based Subquadratic came out of stealth claiming it cracked the quadratic scaling problem that’s been a fundamental constraint on transformer-based LLMs for nearly a decade — and now it’s starting to show actual receipts. MIT Tech Review notes the initial announcement was met with widespread skepticism, which is the correct response to any stealth-mode claim that you’ve solved something the entire field has been stuck on. But if the math checks out, this is the kind of architectural shift that could quietly reshape the cost and capability ceiling of every model running today.
OpenAI’s AI Diagnosed 18 Rare Childhood Diseases That Had Stumped Doctors
Researchers using an OpenAI reasoning model identified 18 new diagnoses in previously unsolved rare pediatric genetic disease cases — children who had gone undiagnosed through conventional medicine. This is the kind of AI application that cuts through all the hype and hits at something real: rare diseases are brutally hard to diagnose precisely because no single physician can hold the entire pattern-space in their head, and AI can. More of this, please.
OpenAI Launches LifeSciBench to Measure Real Scientific Reasoning
OpenAI introduced LifeSciBench, a new benchmark built by domain experts to evaluate how AI handles actual life science research tasks — not toy problems, but the messy, judgment-heavy decisions that real researchers face. Benchmarks are only as good as what they measure, and the history of AI benchmarks being gamed or outgrown within months is not encouraging, but expert-authored evaluations targeting genuine research decisions are at least pointed in the right direction. The real test will be whether it stays ahead of the models long enough to be meaningful.
The UK Is Using Flawed Facial Recognition on Asylum Seekers Anyway
Wired reports that the UK Home Office’s own internal tests showed significant error rates in facial age-verification technology — and the government is deploying it on asylum seekers anyway. The consequences of getting it wrong aren’t minor; they’re life-altering determinations about a person’s legal status. This is the part of the AI policy conversation that gets less attention than IPOs and benchmarks: the quiet deployment of imperfect systems on the people least positioned to push back.
Brain-Computer Interface Trials Are Accelerating — and Changing Lives
MIT Tech Review profiles Casey Harrell, an ALS patient who’s spent nearly three years as what researchers call “the first power user” of a brain-computer interface — a device that’s restored his ability to communicate despite being paralyzed and unable to speak clearly without it. BCI trials are proliferating, and the distance between “promising research” and “technology that gave a man his voice back” is getting shorter. This one deserves more attention than it’s getting.
Bottom Line
The most important AI stories this week aren’t about who’s raising a billion dollars — they’re about a government banning technology it doesn’t understand, a disease diagnosed in a child who’d run out of other options, and a paralyzed man talking to his family again.