░▒▓██▓▒░
░▒▓██████▓▒░
░▒▓████████████▓▒░
░▒▓██████████████████▓▒░
T H E U N C E R T A I N F U T U R E S
all the news that fits the context window
──────────────────────────────────────────
No. 13 · Fri · 26 June 2026
Happy Friday, and thanks to everyone who read the newsletter this week — Monday's blue, Tuesday's grey, Wednesday too, but it's Friday now. Washington spent the week deciding which AI models can ship, asking OpenAI to stagger GPT-5.6 and leaning on Meta to submit its models for review. Anthropic says it caught Alibaba copying Claude at industrial scale, and Apple passed the memory crunch straight to its customers — pricier Macs, and a stock that slid on the news.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░
Washington moves to gate which AI models can ship — it asks OpenAI to stagger GPT-5.6 and presses Meta into model reviews
The Trump administration asked OpenAI to release its next model, GPT-5.6, only to a small set of government-approved partners first, with federal officials signing off on preview access "customer by customer," The Information reported. The White House is separately pressing Meta to enter the same voluntary testing program at the Commerce Department's AI standards center — OpenAI, Microsoft, Google, and xAI have already joined — which would leave Meta the last major lab outside it. It is the first time Washington has told an American lab to hold a model back before launch, and it extends the control it began when it pulled Anthropic's flagship Fable model offline this month. Covered by The Information, NYT, and Axios. (The Information · Axios · NYT)
Anthropic accuses Alibaba of the largest theft of Claude's abilities it has caught
Anthropic told US senators that Alibaba and its Qwen AI-model lab ran more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts between April 22 and June 5, harvesting its answers to train cheaper models — a technique called distillation that sidesteps the export controls on model weights. In a letter dated June 10 and reported this week, the company called it "the largest known distillation attack" it has detected to date, and asked Congress to penalize foreign labs that do it and to tighten chip controls. Alibaba declined to comment; the accusation lands as US officials say aloud that China's AI gap is closing. Covered by Reuters, Bloomberg, BBC, The Information, and GZERO. (BBC · Reuters)
Apple raises Mac and iPad prices and its stock falls about 6% as the AI memory crunch reaches shoppers (update)
Apple raised prices on MacBooks and iPads on Thursday, saying it could no longer absorb the cost of memory and storage chips that AI data centers are buying up — the 14-inch MacBook Pro climbs from $1,700 to $2,000, with increases estimated at 15% to 25% across the lineup. The shares fell about 6%, Apple's worst day in more than a year, as investors weighed whether higher prices choke demand; the iPhone was spared, for now. Microsoft raised Xbox prices the same day, a sign the squeeze the brief flagged last week is now a consumer-wide event, not an Apple problem. Covered by Reuters, Bloomberg, and The Information. (Investing.com / Reuters · Bloomberg)
Qualcomm buys software startup Modular for about $3.9 billion to chip at Nvidia's biggest moat
Qualcomm agreed to pay roughly $3.9 billion in stock for Modular, whose software lets AI models run across different chips without rewriting the code — a direct swipe at CUDA, the Nvidia software layer that keeps developers tied to Nvidia hardware. The all-stock deal more than doubles Modular's $1.6 billion valuation from September and brings in co-founder Chris Lattner, who created Apple's Swift language. Qualcomm makes most of its money on phone chips and has been hunting for a way into the AI data center; buying the software layer is a cheaper route than out-engineering Nvidia's silicon. Covered by The Information, WIRED, and Quartz. (Quartz · The Information)
Google keeps losing AI talent to Anthropic and pushes back its flagship Gemini model (update)
Two more senior DeepMind researchers, Jonas Adler and Alexander Pritzel, are leaving for Anthropic, following a Nobel laureate who left for the same lab earlier this month. Google has now delayed Gemini 3.5 Pro to July to keep stress-testing it on real coding work, and is reshuffling an internal coding strike team to close the gap with Anthropic's Claude. The throughline: the company that wrote the transformer paper is now the one playing catch-up, paying London-built talent to walk a few miles to Anthropic's new office. Covered by Bloomberg, TechCrunch, and Business Insider. (TechCrunch · Business Insider)
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░
░░▒▒▓▓ AI ▓▓▒▒░░
░░▒▒▓▓ POLICY ▓▓▒▒░░
░░▒▒▓▓ INFRA ▓▓▒▒░░
░░▒▒▓▓ BUSINESS ▓▓▒▒░░
░░▒▒▓▓ THE WORLD ▓▓▒▒░░
░░▒▒▓▓ ALPS ▓▓▒▒░░
░░▒▒▓▓ RESEARCH ▓▓▒▒░░
░░▒▒▓▓ ROBOTICS ▓▓▒▒░░
░░▒▒▓▓ REPORTS ▓▓▒▒░░
░░▒▒▓▓ COMMS DESK ▓▓▒▒░░
Chamath Palihapitiya, an early Facebook executive, said on "The Axios Show" that the company "profoundly failed" in AI and "completely fumbled" a lead its trove of user data should have guaranteed, pointing to Nvidia — not Meta — as the one that recognized the open-weight moment (Axios). Takeaway: a specific charge from a founding insider — "completely fumbled" — is far harder to answer than a vague one, and a roadmap will not do it; the only credible reply is product the audience can already try.
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░
A Google engineer, Justin Poehnelt, says he was fired this spring for building a tool that made Google's own services easier to use: an open-source command-line interface for Gmail, Drive, and Calendar that hit number one on Hacker News and drew thousands of users in days. The detail that went viral with it was the timing — Google announced its own official Workspace command-line tool two days before letting him go. The surest way to kill a useful side project is to make it popular. (Office Chai)
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░
🎵 The Nightmare [1] — Two Shell
Im Schatten — your editors at The Uncertain Futures