The Uncertain Futures — 2026-06-10

░▒▓▓▒░  ░▒▓▓▒░  ░▒▓▓▒░  ░▒▓▓▒░  ░▒▓▓▒░
╔════════════════════════════════════════╗
║         THE UNCERTAIN FUTURES          ║
╚════════════════════════════════════════╝
No. 1 · Wednesday · June 10, 2026
all the news that fits the context window

Welcome to the first edition. It lands on a heavy news day even by 2026 standards: the most capable model Anthropic has ever sold to the public went on sale, the company behind ChatGPT filed to go public, and Washington spent the week working out how much of both it wants to own.

░░▒▒▓▓  TOP 5  ▓▓▒▒░░

Anthropic releases Claude Fable 5, the first Mythos-class model open to the public

Fable 5 went on sale Tuesday through the Claude API at $10 per million input tokens and $50 per million output — less than half the Mythos Preview rate — with classifier safeguards that hand cybersecurity, biology-chemistry, and model-distillation requests down to Claude Opus 4.8, a fallback Anthropic says triggers in under 5% of sessions. All Mythos-class traffic now carries a mandatory 30-day data-retention policy for jailbreak detection, while the unrestricted sibling, Claude Mythos 5, goes only to vetted cyberdefenders as Project Glasswing expands to about 150 organizations. Early testers describe a wide capability gap over every public model: Ethan Mollick reports Fable working unattended for 9.5 hours to deliver a research-calibration tool academics have wanted for years. Covered by StrictlyVC, Don't Worry About the Vase, One Useful Thing, and Platformer. Primary: anthropic.com

OpenAI files confidentially for IPO and declares its "third phase"

OpenAI confirmed Monday it has submitted a confidential S-1 — one week after Anthropic's — while insisting timing is undecided ("it may be a while"). The same day, Sam Altman and Jakub Pachocki published "Built to benefit everyone," committing to an automated AI researcher producing "a significant fraction of our research" by March 2028, a personal AGI for everyone on Earth, and an international body empowered to slow frontier development when needed — the same coordination call Anthropic made a week earlier. Covered by AI Daily Brief, The Rundown, and Don't Worry About the Vase. Primary: S-1 statement · the plan

Trump's NSPM-11 locks frontier AI into warfighting — as the NSA runs Mythos

The June 5 memorandum directs military and intelligence agencies to accelerate AI adoption, bars any vendor from disabling or degrading deployed systems without federal approval, and orders termination of contracts — subcontracts included — with companies showing a "pattern of conduct" inconsistent with the policy, language widely read as written for Anthropic after its Pentagon standoff and March "supply chain risk" designation. Days earlier, the FT reported roughly half a dozen Anthropic engineers are embedded at the NSA adapting Mythos for offensive cyber operations — under an explicit carve-out from the same ban the memorandum now codifies. Covered by Don't Worry About the Vase. Primary: whitehouse.gov · TechCrunch on the FT scoop

SpaceX closes its books on $250 billion in demand for Friday's record IPO

Order books for the $75 billion raise close Wednesday nearly four times oversubscribed, with Nasdaq trading set to begin Friday at a target $135 per share and up to 30% of shares reserved for retail platforms. Musk spent the week putting engineering behind the prospectus's $23 trillion space-computing market claim, presenting a satellite design carrying roughly 150 kilowatts of AI compute — about one rack of NVIDIA Blackwells — and targeting one gigawatt of annualized orbital capacity by end-2027. Covered by Reuters Japan, AI Daily Brief, and StrictlyVC. Primary: Bloomberg

Apple ships the Siri it promised in 2024 — and Brussels rejects its excuse for skipping Europe

At WWDC, Apple rebranded its assistant Siri AI, built on custom Gemini-based models under a deal worth about $1 billion a year to Google, adding on-screen awareness, cross-app actions, and a dedicated app — free this fall on iPhone 15 Pro and newer, but absent from the EU and China at launch. The European Commission disputed Apple's claim that the Digital Markets Act forced that gap: "The decision not to roll out Siri AI in the EU is Apple's and Apple's only," said spokesperson Thomas Regnier, after the company was denied an 18-month interoperability exemption. Covered by The Rundown, AI Daily Brief, and StrictlyVC. Primary: TechCrunch · Reuters via RTE

░░▒▒▓▓  STATS OF THE DAY  ▓▓▒▒░░

• $250 billion — investor demand for SpaceX's $75 billion IPO, nearly four times oversubscribed (Bloomberg/Reuters)
• $10 / $50 — Claude Fable 5 pricing per million input/output tokens, less than half the Mythos Preview rate (Anthropic)
• $920 million — Google's monthly compute bill to SpaceX for 110,000+ NVIDIA GPUs through June 2029 (The Information/CNBC)
• 3 million — TPUs Google ordered from Intel for 2028 delivery, Intel's first major AI-era foundry order (The Information)
• £1.1 billion — the UK's new AI Hardware Plan, £750 million of it for a national supercomputer (gov.uk)

░░▒▒▓▓  AI  ▓▓▒▒░░

• OpenAI is rebuilding ChatGPT into an agent-and-coding "superapp" that merges Codex and the Atlas browser; Codex has grown sixfold since February to 5 million weekly users, and a senior employee told the FT "Chat is dead" (FT, via The Rundown and AI Daily Brief).
• Molly Kinder, who led Brookings' work on generative AI and jobs, is leaving to found an organization for the "messy middle" of the AI transition — her bet: most jobs survive, but losses concentrate in the best-paid knowledge work, with politically explosive results (Platformer interview).
• Clive Chan, the second hire on OpenAI's custom-chip team, joined Anthropic — a pointed pickup given Anthropic is weighing building its own silicon (The Rundown).
• Google's NotebookLM update gives each notebook agentic chat and a sandboxed cloud computer that writes and runs Python, plus spreadsheet, slide, and PDF outputs (Google, via The Rundown).

░░▒▒▓▓  TECH  ▓▓▒▒░░

• EU antitrust regulators ordered Meta to restore rival AI assistants' free access to the WhatsApp Business API within five working days — the Commission's first interim measure in 17 years, imposed before the probe concludes (WSJ, via StrictlyVC).
• Meta will begin using activity from other websites and apps — purchases included — to personalize Facebook, Instagram, and Meta AI responses starting in July; the EU and UK are excluded at launch (The Verge, via StrictlyVC).
• Kalshi will require employer disclosure for trades in insider-risk markets such as corporate performance and national security, after referring more than 20 cases to regulators in Q1 (WSJ, via StrictlyVC).
• US campuses now offer at least 74 AI majors and 89 minors, up from five majors in 2021; employer appetite for the credentials remains untested (NYT, via StrictlyVC).

░░▒▒▓▓  POLICY  ▓▓▒▒░░

• President Trump confirmed talks on the US government taking an equity stake in OpenAI — donated rather than purchased — to seed a Public Wealth Fund paying AI dividends to households; Bernie Sanders' rival bill would take 50% of frontier-lab stock through a one-time tax (CNBC/Axios, via The Rundown and AI Daily Brief).
• The White House and Sen. Marsha Blackburn are negotiating a federal package bundling the Kids Online Safety Act, the NO FAKES Act, and age verification — the price for subject-matter preemption of state AI laws (Axios, via AI Daily Brief).
• Sen. Adam Schiff introduced the HALO Act, requiring a designated commander's authority over autonomous-weapons decisions and barring Pentagon AI surveillance of constitutionally protected activity; he wants it attached to the must-pass defense bill (WSJ/The Hill, via AI Daily Brief).
• Argentina sent Congress a bill creating the "non-human corporation" — companies run entirely by AI agents, with liability protection and low taxes; Milei pitched deregulated AI in an FT op-ed, and Yuval Noah Harari answered that AI legal personhood invites an unregulatable "AI state" (Buenos Aires Times/FT, via The Rundown).

░░▒▒▓▓  INFRA  ▓▓▒▒░░

• Seattle's city council voted 9–0 for a one-year emergency moratorium on data centers above 20 megavolt-amperes; more than 50 residents testified, none in favor of the industry (GeekWire).
• Google placed Intel's first major AI-era foundry order — over 3 million TPUs for 2028 — and Nvidia is evaluating Intel's 18A process, as TSMC warns demand will outrun supply for years; Intel jumped 13% premarket (The Information, via AI Daily Brief and The Rundown).
• The UK's £1.1 billion AI Hardware Plan routes £750 million to a national supercomputer at Edinburgh and £150 million in advance purchase commitments for chips from British startups (gov.uk, via The Rundown).
• China switched on what it calls the world's first wind-powered underwater data center off Shanghai: 24 MW, about 2,000 servers, 97% of power from offshore wind, seawater cooling (The Guardian, via StrictlyVC).

░░▒▒▓▓  BUSINESS  ▓▓▒▒░░

• Google will pay SpaceX $920 million a month from October through June 2029 for at least 110,000 Nvidia GPUs at xAI-built data centers — SpaceX's second mega-customer after Anthropic took all of Colossus 1 in May (The Information/CNBC, via AI Daily Brief).
• Nvidia and SK Hynix signed a multiyear memory partnership spanning Vera Rubin supercomputers to robotics platforms, securing HBM supply; Jensen Huang: "Everything in the entire industrial supply chain... is in a state of supply shortage" (Reuters/NVIDIA, via AI Daily Brief).
• Goldman Sachs and JPMorgan are exploring futures contracts tied to GPU rental prices, and CME Group plans a compute-futures market later this year — the financialization of the AI buildout's biggest cost line (The Information, via AI Daily Brief).
• Moonshot AI, maker of the Kimi models, is raising $1–2 billion at up to a $30 billion valuation — seven times its December mark, on annual recurring revenue that passed $200 million in April (SCMP/Bloomberg, via The Rundown).

░░▒▒▓▓  GEOPOLITICS  ▓▓▒▒░░

• Xi Jinping made his first Pyongyang visit since 2019, affirming North Korea's "sovereignty and security" interests without once mentioning denuclearization — days after Kim unveiled a third enrichment site and claimed doubled warhead-production capacity since 2021 (ISW).
• Ukraine's AI-enabled mid-range drone campaign now reaches 150 miles behind Russian lines; Moscow has banned military cargo on the main Rostov–Crimea highway and is reportedly pulling depleted units off the Kinburn Spit as resupply collapses (ISW, The War Zone).
• France and Germany scrapped FCAS, Europe's $116 billion sixth-generation fighter program, after Dassault and Airbus deadlocked over workshare — Airbus is already courting Saab (GZERO).

░░▒▒▓▓  REPORTS  ▓▓▒▒░░

McKinsey — When AI becomes part of the workflow: Redesigning how software gets built. Case study built around code-quality firm Sonar: the gains come from redesigning the whole product-development life cycle, not bolting AI onto one stage — top performers are six to seven times likelier to scale AI across four or more use cases, and CEO Tariq Shaukat warns against measuring output in lines of code rather than what the code achieves. Report page

░░▒▒▓▓  COMMS DESK  ▓▓▒▒░░

OpenAI's "Built to benefit everyone" is the season's cleanest example of multi-audience positioning: one document selling the dream to consumers, the market size to investors, and public benefit to regulators — published the same day as the company's confidential S-1. The framing war is running in parallel, with OpenAI's Joshua Achiam casting the rivalry as "machine God" (Anthropic) versus "tools of progress" (OpenAI), a characterization Anthropic staff publicly reject (AI Daily Brief; Don't Worry About the Vase). Takeaway: when a company narrates its own history in numbered phases, the audience is the market — treat era-declarations as positioning artifacts, not roadmaps.

Palantir co-founder Joe Lonsdale accused CEOs of citing "AI productivity" to dress up layoffs actually caused by 2021–2023 overhiring or missed growth, naming Block, Atlassian, and Coinbase; Marc Andreessen concurred (Business Insider, via StrictlyVC). Takeaway: the AI-layoff narrative is curdling fast — if leadership reaches for it, demand receipts before it goes anywhere near a press release.

AI Signal's framework piece argues AI-assisted content fails when one draft gets reshaped per platform instead of re-translated per audience: hold the point of view constant, let hook, structure, and rhythm move (AI Signal). Takeaway: brief the model on what must not change, not only on what to produce.

░░▒▒▓▓  ONE MORE THING  ▓▓▒▒░░

Catherine Breed, a 33-year-old Bay Area open-water swimmer, has set off on the first attempt to swim California's entire 900-mile coast — Oregon to Mexico, five-plus hours a day for up to four months, sleeping aboard a support boat. She names sharks and elephant seals as her chief concerns, which suggests the jellyfish, storms, and hypothermia have been filed under manageable (NYT, via StrictlyVC).

░░▒▒▓▓  TRACK OF THE DAY  ▓▓▒▒░░

🎵 Tranki — Skrillex, Tracey, TAICHU, ANITA B QUEEN

En guete — the Uncertain Futures desk