The Uncertain Futures — No. 4 — the last edition as Temperature Zero

╔══════════════════════════════════════╗
║░▒▓ T E M P E R A T U R E  Z E R O ▓▒░║
╠══════════════════════════════════════╣
║ No. 4 · Monday · June 15, 2026       ║
║ ▂▃▄▅▆▇█ █▇▆▅▄▃▂  ▂▃▄▅▆▇█ █▇▆▅▄▃▂     ║
╚══════════════════════════════════════╝

Good Monday, and a warm hello to everyone who signed up over the weekend. The Fable shutdown turned up its source, SpaceX closed its first public day worth more than Broadcom, and Meta began rationing the very AI tools it spent a year ordering everyone to use. Welcome to Temperature Zero.

░░▒▒▓▓  TOP 5  ▓▓▒▒░░

The US order that pulled Fable 5 traces back to Amazon, and to CEO Andy Jassy, the Wall Street Journal reports (update)

The export-control directive that took Anthropic's Fable 5 and Mythos 5 dark worldwide on Friday, as our weekend extra covered, now has a name attached to it. The Wall Street Journal reports that Amazon researchers found the jailbreak and that CEO Andy Jassy carried the alarm to Washington, after which Commerce Secretary Howard Lutnick sent Dario Amodei the letter. The awkward part: Amazon is at once a $13-billion investor in Anthropic and a Project Glasswing security partner contractually meant to route findings through that channel rather than to the federal government. The European Commission, for its part, says it is still working out what a global cutoff means for European users. Covered by The Information, TechCrunch, Reuters, Exponential View, and AI Daily Brief. Primary: WSJ · TechCrunch

SpaceX closed its first public day up 19% at $160.95, then promptly landed on governance funds' do-not-buy lists (update)

SpaceX opened on Nasdaq at $150 and settled Friday at $160.95, 19% above the $135 IPO price, the sixth-most-valuable US public company on day one at roughly $1.77 trillion; it cleared $2 trillion intraday, briefly passing Broadcom. The $75 billion raise broke Saudi Aramco's 2019 record and lifted Elon Musk's paper net worth past $1 trillion. Not everyone was buying: Bloomberg reported governance-mandated funds calling the stock "simply too risky," Denmark's Akademikerpension added it to its exclusion list over Musk's near-total voting control, and S&P's refusal to fast-track the listing keeps it out of the S&P 500, and the passive money that follows it, for at least a year. Covered by CNBC, Reuters, Bloomberg, The Information, and NZZ. Primary: CNBC · Bloomberg

Meta caps employee AI spending after staff ran up 73.7 trillion tokens in a month

Meta told staff in a June 12 memo it will impose per-team limits on AI token use, after internal forecasts ran into the billions and a usage leaderboard logged 73.7 trillion tokens in roughly 30 days. Chief technology officer Andrew Bosworth wrote the memo, walking back his own earlier rallying cry to staff: "Keep doing it. No limit." Mark Zuckerberg told Reuters the company had made mistakes in its AI hiring spree, and Uber and Amazon have made the same move as the invoices arrive. The strain runs deeper than budgets: WIRED reports Meta's three-month-old, 6,500-person Applied AI unit is near revolt, with one all-hands ending in an employee calling a senior Meta AI executive "a piece of sh*t." Covered by The Information, Reuters, GeekWire, WIRED, and Business Insider. Primary: The Information · Reuters

Oracle's slide deepened to 11% as the market fixed on a $23.7 billion cash-flow hole (update)

Oracle's post-earnings slide deepened to 11%, down from the 7% after-hours dip when results landed last week, as attention moved from the headline beat to the bill behind it: fiscal-2026 free cash flow of negative $23.7 billion against capital spending up 162% to $55.7 billion, with plans to raise about $40 billion in debt and equity and fiscal-2027 capex guided near $70 billion. The AI buildout is real; so is the question of when it starts paying for itself. Covered by CNBC and TLDR. Primary: CNBC

The AI price war the labs warned about is here, and enterprises are routing real work to DeepSeek and Alibaba's Qwen

The boom is turning into a price war, the Wall Street Journal reported June 12: companies are cutting costs by routing workloads to cheaper open-source and Chinese models, with DeepSeek and Alibaba's Qwen the names that keep recurring as the capability gap with the frontier narrows. OpenAI is weighing drastic price cuts of its own. The squeeze lands on the two labs least able to shrug it off, OpenAI and Anthropic, both losing billions on compute while marching toward IPOs, which makes it the strategic backdrop to much of the rest of this week. Covered by StrictlyVC, AI Daily Brief, and Exponential View. Primary: WSJ

░░▒▒▓▓  STATS OF THE DAY  ▓▓▒▒░░

• $160.95 — SpaceX's first-day close, up 19% on its $135 IPO price (CNBC)
• 73.7 trillion — tokens Meta staff burned in about 30 days before the caps landed (The Information)
• −$23.7 billion — Oracle's fiscal-2026 free cash flow, even as capex rose 162% to $55.7 billion (CNBC)
• 206 — vulnerabilities in Microsoft's record June patch, its biggest ever (CyberScoop)
• $1.77 trillion — SpaceX's day-one market value, sixth-largest US public company (CNBC)

░░▒▒▓▓  TECH  ▓▓▒▒░░

• A flaw disclosed in June in Meta's AI customer-service chatbot let attackers reset passwords on roughly 20,000 Instagram accounts by feeding it any email the system failed to verify; the marks included a former White House account for Barack Obama, a US Space Force official, and Sephora. The support bot, it turned out, supported the attackers (The Median).
• Microsoft has weighed spinning Xbox out as a separate business, The Information reported June 12, with new gaming chief Asha Sharma instead pushing a turnaround built on accelerated Halo, Fallout, and Elder Scrolls releases, franchises that have gone 5 to 15 years between mainline titles (The Information · GameSpot).
• Further on Microsoft, the company patched a record 206 vulnerabilities this month, its largest batch since Patch Tuesday began in 2003, with three of them zero-days. Security researchers tie part of the surge to AI-assisted vulnerability discovery, and one open reading connects it to the week's bigger story: the same frontier-model capability that got Mythos pulled offline, finding security flaws at scale, is what keeps inflating every vendor's patch count, Microsoft's included (CyberScoop).

░░▒▒▓▓  POLICY  ▓▓▒▒░░

• US warrantless-surveillance powers under Section 702 lapsed on June 12 for the first time, after the Senate rejected President Trump's pick to lead the intelligence agencies, leaving NSA and FBI collection authority expired while the fight plays out (NPR).

░░▒▒▓▓  BUSINESS  ▓▓▒▒░░

• Kioxia, the Japanese flash-memory maker spun out of Toshiba, overtook Toyota on June 12 to become Japan's most valuable listed company, a tidy measure of how far the AI-memory boom has rearranged the market when a chipmaker outweighs the carmaker that defined the country (Bloomberg).
• The Information reports Anthropic is rolling out apps that compete with its own enterprise customers: Figma walked away from a Claude Design launch partnership in April, accusing the company of deceptive communication, while Claude Code presses on Cursor, a playbook critics liken to Microsoft's and Google's at their most acquisitive (The Information).
• Visa wired its payment network into ChatGPT in June, letting AI agents shop and pay at Visa merchants on a user's behalf, handing the agents a checkout button and the rest of us a new way to wonder what we just bought (AP).
• Sam Bankman-Fried lost his appeal on June 12: a federal court upheld his FTX fraud conviction and 25-year sentence, leaving the US Supreme Court as his last resort (Reuters).

░░▒▒▓▓  GEOPOLITICS  ▓▓▒▒░░

• A US-Iran framework to wind down the three-month war is now set for a formal signing in Geneva on Friday, mediated by Pakistan, though Tehran disputes parts of the draft and the Strait of Hormuz stays closed pending mine-clearing (NZZ).
• Anti-immigrant violence in South Africa has killed at least five foreign nationals and displaced thousands over the past month, with a self-styled movement setting a June 30 deadline for undocumented migrants to leave; Ghana, Nigeria, and Malawi have begun repatriating citizens (GZERO).

░░▒▒▓▓  ALPS  ▓▓▒▒░░

• Mistral is in talks to raise about €3 billion at a roughly €20 billion valuation, Bloomberg reported June 12. Europe's largest AI lab is building the kind of war chest the shutdown of a US frontier model just made more attractive to European buyers (Bloomberg).
• AI-lab executives from Anthropic, OpenAI, Google, and Mistral are expected at this week's G7, where EU tech chief Henna Virkkunen said Brussels will not take equity stakes in AI firms and blamed Apple, not the Digital Markets Act, for withholding Siri's AI from the bloc (Axios · POLITICO).
• The Swiss parliament could lift the country's ban on new nuclear plants on Monday, with the FDP calling it energy security and the Centre warning the bill promises reactors without naming their cost (NZZ).
• EPFL researchers in Lausanne built an AI, reported in June, that works out how to stimulate the brain so vision prostheses render whole objects instead of scattered points of light, with promising early tests in monkeys (kinewsletter.ch).

░░▒▒▓▓  RESEARCH  ▓▓▒▒░░

• Jennifer Doudna's institute published a CRISPR system in Nature this month that reads the RNA signature of a p53-mutant cell and shreds its genetic material while sparing healthy neighbors: a run at a cancer driver that has resisted drugmakers for decades, so far proven in mouse models of lung and liver tumors (Innovative Genomics via Not Boring).
• NOAA declared on June 11 that El Niño has formed in the Pacific and expects it to reach historic strength, rivaling the destructive 1997 event, with several scientists already pencilling in 2027 as the hottest year on record (NOAA).

░░▒▒▓▓  COMMS DESK  ▓▓▒▒░░

The Fable affair is a case study in disclosure timing deciding a story the facts alone would not. A government letter that lands at 5:21 on a Friday with no specifics; an investor-and-security-partner, Amazon, that escalated to Washington rather than through the channel built for exactly this; a company that apologized for one safeguard while defending the rest. Anthropic's fast, plain reversal of the silent throttle drew grudging praise even from critics, but the damage was set the moment the disclosure took a route nobody could control. Takeaway: a partner escalation path only works if everyone actually uses it; the instant someone routes around it, the routing becomes the headline, and you forfeit the chance to frame your own bad news.

Separately, Anthropic's move into apps that compete with its own customers (Figma quit a launch partnership accusing it of deceptive communication) is now a named reputational liability, not a hypothetical one (The Information). Takeaway: if a roadmap will cross a partner's core business, brief them before the reporters do, or lose the relationship and the framing in the same week.

░░▒▒▓▓  ONE MORE THING  ▓▓▒▒░░

In China, WIRED reported this month, $30 buys you a chauffeur: a thumb-sized plastic replica of Dwayne Johnson's head, suction-cupped above the rearview mirror, that fools Tesla's driver-monitoring camera into seeing an attentive driver. One owner says he ran 250 of a 400-mile trip with the fake head in place, one hand on the wheel and the other in a bag of sunflower seeds. The deluxe versions loop a video of a person blinking; the seller advertises a "0% error rate," which is more than Tesla can claim (WIRED).

Bis morgen — the Temperature Zero desk