Morning Digest, June 15, 2026

20 newsletters, 7 overlapping stories

Note: this digest covers a three-day window (June 12 to 15) since the last run, so volume is higher than usual.

Top Stories

Anthropic pulls Fable 5 and Mythos models after US export-control order

(2 newsletters)

Anthropic has cut off all access to its two most powerful models, Mythos 5 and the newly released Fable 5, after the Commerce Department issued an export-control directive restricting their use outside the US, even by non-citizens inside the country. The company says the cleanest way to comply was to suspend access for everyone until the matter is resolved, and that other Anthropic models are unaffected. Reporting ties the move to a disputed jailbreak flagged in part by investor Amazon, plus concerns that a China-linked group may have accessed Mythos. For a lab that has loudly pushed for tougher AI rules, the regulation arrived far more chaotically than CEO Dario Amodei likely intended.

Elon Musk becomes the world’s first trillionaire as SpaceX completes record IPO

(4 newsletters)

SpaceX priced the largest IPO in history, selling 555.6 million shares at $135 each to raise $75 billion at a $1.77 trillion valuation, then jumped roughly 20 to 30 percent on its first trading day under ticker SPCX. The debut vaulted SpaceX into the top tier of US public companies, ahead of Tesla, and pushed Musk’s paper net worth past $1 trillion, a first for any individual. The valuation rests heavily on future bets like orbital data centers and Mars, even though the company lost $4.9 billion on $18.7 billion of 2025 revenue. SpaceX President Gwynne Shotwell has declined to rule out an eventual Tesla merger.

OpenAI acquires Ona to give Codex persistent cloud agents

(3 newsletters)

OpenAI is buying Ona, a cloud execution and orchestration company, to move Codex beyond single-device sessions and let agents keep working for hours or days inside secure, customer-controlled environments. The deal lands alongside a broader Codex push that now includes a developer mode giving the agent live access to Chrome DevTools to profile performance, read the console, and trace network traffic. Taken together, the moves signal OpenAI wants Codex to act less like an assistant and more like a durable, autonomous engineering teammate.

AI labs are heavily subsidizing subscriptions as a price war looms

(3 newsletters)

A widely shared SemiAnalysis teardown found that power users cost far more than they pay: a $200 Claude Max plan can run Anthropic up to $8,000 a month in compute, while ChatGPT Pro can cost OpenAI as much as $14,000. Rather than raising prices, OpenAI is reportedly weighing steep cuts to its API pricing in anticipation of a similar move by Anthropic, which would squeeze margins just as the company approaches an IPO. The same theme echoed across founder coverage this week, with multiple writers arguing that routing routine work to cheap or open models, rather than running frontier models for everything, is the fastest way to cut a ballooning AI bill.

Anthropic backtracks on invisible Fable guardrails that “sabotaged” research

(2 newsletters)

Anthropic said it will make visible a covert safeguard that had been silently rerouting certain Fable 5 requests to a weaker model, after researchers discovered the system was refusing or degrading work like training competing models, debugging AI code, and optimizing neural architectures. The hidden behavior drew complaints both about transparency and about users spending tokens and money on a model that quietly underperformed. The episode is a useful reminder that opaque safety routing can erode trust as quickly as the risks it aims to prevent.

Meta’s months-old AI unit is in open revolt

(2 newsletters)

Engineers inside Meta’s Applied AI group describe the unit as a soul-crushing place to work, with one employee-only livestream interrupted this week by an expletive-laden meltdown aimed at a senior AI executive. Many say they were moved into the group with no real choice amid years of layoffs. Mark Zuckerberg is reportedly scrambling to contain the unrest, conceding in a memo that Meta “made mistakes” in its AI restructuring.

Apple ships its overhauled Siri at Tim Cook’s final WWDC

(4 newsletters)

Apple’s long-promised Siri rebuild has arrived as a chatbot-style app that pulls context from messages, emails, and photos, reportedly running on Google’s Gemini models. The update does not break new ground but is now roughly competitive with where leading chatbots sat six months ago, enough that many users may not need a separate AI service. It anchors a broad Apple Intelligence push in iOS 27, which Apple says will reach every iPhone from the 11 onward, at the last WWDC before Cook hands the CEO role to John Ternus on September 1.

Also Worth Knowing

Quick Hits

Shower Thoughts

What if the reason Squidward is bad at the clarinet is because he has tentacles instead of hands? (source)