░▒▓█▓▒░ ▒▓██▓▒ ░▒▓█▓▒░ ▒▓█▓▒
▒▓███▓▒▒▓█████▓▒░▒▓███▓▒▒▓███▓▒
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. 7 · Thu · 18 June 2026Good Thursday, and a warm hello to everyone reading today. Anthropic is having the strangest month in AI — pulled offline by Washington one week, outselling OpenAI to businesses the next; the man whose paper started the whole boom just walked from Google to OpenAI; and the SpaceX rocket finally met a Federal Reserve that doesn't care how high it flew. Let's get into it.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░Anthropic overtook OpenAI in business adoption for the first time
Anthropic pulled ahead of OpenAI in business-adoption share for the first time in April, at 34.4 percent of companies tracked versus OpenAI's 32.3 percent, per card-platform Ramp's AI Index — a share that stood at 9 percent a year ago. That lead predates Friday's shutdown of its newest model, so the figures can't yet show the ban's effect; but Ramp lead economist Ara Kharazian told TechCrunch the fight may help rather than hurt, since it appears to validate the fuss over how capable the models are. Anthropic also raised $65 billion at a $965 billion valuation in late May and filed confidentially to go public, on the strength of what it says was its first profitable quarter. Covered by TechCrunch, StrictlyVC, and Ramp. Primary: TechCrunch · Ramp
Noam Shazeer leaves Google to join OpenAI, less than two years after Google paid $2.7 billion to get him back
OpenAI told staff on Wednesday that Noam Shazeer — a co-author of the 2017 Transformer paper that underlies modern generative AI, and most recently co-lead of Google's Gemini — is joining the company. Google had paid roughly $2.7 billion in 2024 to license his startup Character.AI and bring him back to run Gemini; he is now leaving Google less than two years later, as OpenAI staffs up ahead of a likely IPO and the competition for senior researchers turns feverish. Rarely does a named architect of the field cross straight from one frontier lab to its closest rival. Covered by The Information and CNBC. Primary: CNBC · The Information
Tech stocks sell off as the Fed signals a rate hike; SpaceX drops ~5% and falls back below Amazon
US stocks fell on Wednesday after the Federal Reserve held rates steady but signaled an increase could come later this year — new chair Kevin Warsh's first meeting — and the Nasdaq closed off about 1.3 percent. SpaceX, which had touched $225 intraday on Tuesday and briefly passed both Amazon and Microsoft, dropped roughly 5 percent on Wednesday to close near $192, its first full-session loss since listing, slipping back below Amazon. Software stocks took it on the chin: Salesforce is now down 41 percent year-to-date, ServiceNow 38 percent, and Microsoft 21.6 percent, by far the worst of the big-tech group. Covered by The Information, Bloomberg, and Reuters. Primary: TheStreet · Bloomberg
Washington tells Britain no on an Anthropic carve-out, and floats a "trusted partners" scheme instead (update)
The standoff over Anthropic's banned Fable 5 and Mythos 5 models hardened on two fronts on Wednesday. UK Prime Minister Keir Starmer asked Trump at the G7 in France for an exemption restoring access for British citizens and firms, and was refused — granting a carve-out for any ally, even a G7 one, would be "completely illogical," an administration official told the New York Post. In parallel, G7 leaders said they discussed a "trusted partners" scheme that would grant select non-US nations access to top American models, while administration officials, WIRED reports, now treat fixing the jailbreak as Anthropic's problem after the NSA concluded the model's guardrails can be disabled. Covered by Reuters, StrictlyVC, WIRED, and The Rundown. Primary: Reuters · The Next Web
The Pentagon says it used xAI's Grok to plan the Iran bombing — 2,000 munitions on 2,000 targets in 96 hours (update)
A court filing from the Pentagon's chief digital and AI officer, Cameron Stanley, says a Grok model deployed inside Palantir's Maven Smart System — the targeting platform Palantir builds for the Pentagon — supported targeting at the start of the February campaign against Iran, enabling US forces to "deploy over 2,000 munitions to 2,000 distinct targets within 96 hours." The declaration came as the Justice Department pressed the bid we covered Wednesday to dismiss the NAACP suit over the unpermitted gas turbines powering xAI's Mississippi data center, again arguing the plant is vital to national security. The choice of Grok is pointed: Maven first ran on Anthropic's Claude, until Anthropic and the administration fell out earlier this year over the lab's refusal to loosen a usage policy that bars its models from powering autonomous weapons that target or kill without a human in the loop. So Anthropic — the lab Washington pulled offline on Friday — is the one that drew a hard line against exactly this use, even as the same administration went to court this week to shield xAI's Mississippi data center as a strategic asset. Covered by The Information, WIRED, and AI Daily Brief. Primary: The Information · WIRED · Anthropic
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░░░▒▒▓▓ AI ▓▓▒▒░░░░▒▒▓▓ TECH ▓▓▒▒░░░░▒▒▓▓ POLICY ▓▓▒▒░░░░▒▒▓▓ INFRA ▓▓▒▒░░░░▒▒▓▓ BUSINESS ▓▓▒▒░░░░▒▒▓▓ GEOPOLITICS ▓▓▒▒░░░░▒▒▓▓ ALPS ▓▓▒▒░░░░▒▒▓▓ RESEARCH ▓▓▒▒░░░░▒▒▓▓ ROBOTICS ▓▓▒▒░░░░▒▒▓▓ REPORTS ▓▓▒▒░░░░▒▒▓▓ COMMS DESK ▓▓▒▒░░OpenAI and Anthropic have hired close to 90 Salesforce staff this year, mostly into sales and marketing, as the two labs poach enterprise go-to-market muscle for the business customers they are now fighting over (The Information · GeekWire). Separately, "tokenminimizing" has become the phrase of the season: AT&T has begun throttling some employees' AI usage as agentic tools rack up real per-task model bills, WIRED and The Information report (The Information). Takeaway: the comms-leadership lesson in both is the same — the AI line item is moving from free productivity to a budget your CFO will ask you to defend, so price the tokens into any agent rollout before you pitch it as a saving.
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░A leak has exposed the membership of "Dialog," Peter Thiel's invitation-only retreat for the global elite, where, per WIRED, more than 200 attendees registered for an agenda running from panels on cult-building and sex to prepping for World War III — with an associated app that offers, among other things, matchmaking. For a society built on discretion, the one feature it could not resist was a directory. (The Print)
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░Pfiäti — your editors at the Uncertain Futures