░▒▓▓ THE UNCERTAIN FUTURES ▓▓▒░
────────────────────────────────────────
No. 5 · Tue · 16 June 2026From the desk: We have a new name! You signed up for Temperature Zero. It turns out another AI brief already has that brand occupied, so to spare everyone the confusion, we're now The Uncertain Futures. The future's uncertain enough as it is; your daily morning brief, mercifully, is not.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░Anthropic lobbies to reverse the Fable ban as the case for it frays — the "jailbreak" was the model reviewing code (update)
The export-control order that took Fable 5 and Mythos 5 dark on Friday is still in force, and since then the order has lost ground on both fronts — the politics and the technical case alike. Anthropic sent its top security staff — Nicholas Carlini, model-risk lead Logan Graham, and safeguards head Dave Orr — to meet Commerce officials in person, while an open letter signed by more than 70 cybersecurity leaders and organized by former Facebook security chief Alex Stamos asked Commerce Secretary Howard Lutnick to lift the controls. The two sides tell different stories about how it came to this: Anthropic says it got a 90-minute deadline with no specifics, while a senior White House official told Politico the controls were a last resort after hours spent asking Anthropic to pull the models voluntarily — with Treasury Secretary Scott Bessent telling Dario Amodei that refusing to do so was a bad decision. The trigger itself now looks thin: Katie Moussouris of Luta Security, apparently the only outside expert to have read Amazon's paper, says Amazon merely fed Fable buggy open-source code and asked it to find the flaws — the model working as designed, not a hole to patch — and that the guardrails were only ever meant to block working exploits, which Amazon never demonstrated. To fix what Amazon described, the awkward implication runs, Anthropic would have to disable Fable's ability to write code at all. Amazon declined to discuss the details. Covered by AI Daily Brief, Understanding AI, TechCrunch, Axios, GeekWire, The Information, Stratechery, and Where's Your Ed At. Primary: Politico · Luta Security · Axios
Britain moves to bar under-16s from all social media — the widest such ban yet, and it reaches AI companion apps
Prime Minister Keir Starmer said on June 15 that the UK will keep children under 16 off every major platform — Instagram, Facebook, Snapchat, TikTok, YouTube, and X — going, by his own account, further than any country in the world, past even Australia's pioneering law. The plan also bars under-16s from livestreaming, messaging strangers inside games, and using AI romantic-companion chatbots; the government is also weighing overnight curfews for all under-18s and default-on safety settings for 16- and 17-year-olds. The burden falls on the platforms, which face heavy fines for letting minors slip through. First rules are due before Parliament by year's end, to take effect around spring 2027, building on the existing Online Safety Act — and Elon Musk has already called it a police-state move. Covered by Bloomberg, CBS News, NPR, CBC, and CNBC. Primary: NPR · Bloomberg
A leaked audit puts OpenAI's 2025 loss at $38.5 billion, nearly eight times the prior year, on $34 billion of spending
Ed Zitron, citing audited financials he says the Financial Times independently verified, reports OpenAI lost $38.53 billion in 2025 against $13.07 billion in revenue, with total costs of $34 billion — $19.18 billion of it research and development. The 2024 loss was $5.09 billion, making this nearly an eightfold jump; the year also carried a $41.55 billion fair-value charge tied to the for-profit conversion. OpenAI paid Microsoft $17.2 billion over the year and ended it with about $50 billion in assets, nearly half in cash. "$38.53 billion in losses are astronomical, and far higher than most believed it would be," Zitron wrote, as the company heads toward an IPO. Covered by Where's Your Ed At, The Neuron, and The Information. Primary: Where's Your Ed At
Most of the world now ranks China ahead of the US in AI, days after Trump said America was far ahead
A Public First survey of more than 18,000 people across 15 countries found majorities in 11 of them — including France, the UK, and Canada — now name China, not the United States, as the world's leading AI power; only respondents in the US, Japan, India, and Vietnam still put America first. The finding lands days after President Trump said the US was far ahead, and it rhymes with the week's commercial story — the open Chinese models enterprises keep routing work to on price, DeepSeek and Alibaba's Qwen. Perception is not capability, but it steers the talent and capital that build it. Covered by POLITICO. Primary: IBTimes
Fox agreed to buy Roku for $22 billion, the largest of a one-day wave of AI deals
Fox, the slimmed-down broadcaster run by Lachlan Murdoch, agreed to buy streaming-device maker Roku for $22 billion at $160 a share — nearly double its price a year ago — handing founder Anthony Wood a strong exit and Fox a much larger slice of streaming advertising; Fox shares fell 15% on the price, which works out to roughly 24 times Roku's estimated 2027 cash earnings, against Fox's own multiple of about six. It capped a busy Monday: Salesforce agreed to buy AI customer-service firm Fin, formerly Intercom, for $3.6 billion, and The Information reported Qualcomm in talks to buy AI-chip designer Tenstorrent for $8 to $10 billion. Covered by The Information and Reuters. Primary: The Information
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░░░▒▒▓▓ AI ▓▓▒▒░░░░▒▒▓▓ TECH ▓▓▒▒░░░░▒▒▓▓ POLICY ▓▓▒▒░░░░▒▒▓▓ BUSINESS ▓▓▒▒░░░░▒▒▓▓ GEOPOLITICS ▓▓▒▒░░░░▒▒▓▓ ALPS ▓▓▒▒░░░░▒▒▓▓ RESEARCH ▓▓▒▒░░░░▒▒▓▓ COMMS DESK ▓▓▒▒░░The weekend's most consequential corporate message was not Anthropic's. Microsoft's Satya Nadella posted a careful Sunday note arguing against an AI future in which a few models "eat everything they see" — his case being that a company's real edge is its own learning loop: the institutional knowledge it converts into private evaluations and reinforcement on its internal data, with human expertise growing more valuable, not less, as raw model capability commoditizes. Build, in other words, so you can swap a generalist model out without losing the company's hard-won know-how — a pointed line to draw with Anthropic's own model dark by government order. Microsoft AI's Mustafa Suleyman has separately called Anthropic's models too expensive, a partner and investor putting daylight between itself and a supplier in trouble (Where's Your Ed At · GeekWire). Takeaway: when a partner is in crisis, allies reposition fast and on the record; have your own statement ready before theirs frames you, because silence reads as agreement.
Google chief scientist Jeff Dean told a University of Washington commencement that AI can draft code and summarize data but cannot replace graduates' judgment, ethics, or sense of what is worth building — and drew cheers, where pro-AI commencement speeches have been booed across campuses this spring (GeekWire). Takeaway: the line that lands with a skeptical audience concedes the limits first and claims the upside second; lead with what the tool cannot do.
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░Calm, the meditation app, has added a new way to drift off: a recording of its own lawyers reading the company's terms of service aloud, pitched — without a trace of irony — as a bedtime relaxation track. Axios AI+, which spotted it, had no further notes (Axios AI+).
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░Pour one out for Temperature Zero — your editors, now at the Uncertain Futures