The Information Machine

2026-06-12

The US government issued an export control directive requiring Anthropic to suspend Fable 5 and Mythos 5 access for all foreign nationals on national security grounds, on the same day SpaceX completed a $1.77 trillion IPO justified primarily by AI compute contracts.

What

A US government directive effective June 12 requires Anthropic to block all foreign nationals — including its own foreign national employees — from accessing Fable 5 and Mythos 5, citing national security concerns about a demonstrated jailbreak [1]. Anthropic is complying under legal obligation while publicly disputing the technical basis, arguing the jailbreak amounts to asking the model to read a codebase and fix flaws, a capability it says is widely available from competitors including GPT-5.5 [1]. On the same day, SpaceX priced its IPO at $135 per share and a $1.77 trillion valuation, with investor demand reaching $250 billion against a $75 billion offering; post-pricing coverage confirmed that market participants framed the offering as a bet on AI compute infrastructure rather than a space company [2][3]. An Anthropic-commissioned survey released today found only 15% of Americans trust AI companies to make decisions about AI development — the lowest figure of any institution tested — while 64% cite job loss as their top AI fear and 70% support government involvement in AI regulation [4].

Why it matters

A government directive restricting a frontier model to US nationals only — on undisclosed national security grounds the lab itself contests — sets a template for how AI capabilities may be controlled that extends well beyond the specific jailbreak cited. The SpaceX IPO closing at $1.77 trillion, anchored by AI compute contracts and framed by markets as infrastructure rather than aerospace, confirms that AI infrastructure has become a primary lens for trillion-dollar capital decisions.

Open questions

  • The government has not disclosed the specific national security concern driving the Fable 5/Mythos 5 directive, and Anthropic contests the technical basis, arguing the capability is widely available elsewhere [1] — does this dispute create a legal challenge path, and could it set precedent for how broadly capability-based model restrictions can be applied?

  • SpaceX's IPO closed with $250 billion in demand at a $1.77 trillion valuation, framed explicitly as AI compute infrastructure [2][3] — does this create upward pressure on the pending Anthropic and OpenAI S-1 reviews, given both labs have disclosed major compute contracts as core revenue?

  • Community opposition blocked or delayed 75 US datacenter projects worth $130 billion in Q1 2026 [5], while a counter-analysis on aggregate national water use relies on self-reported company figures and acknowledges local stress [6] — do regulators have independent data sufficient to evaluate local versus national-scale impact claims?

  • Anthropic's own survey found only 15% of Americans trust AI companies on AI decisions, and 70% support government regulation [4] — does this public posture, combined with the export control directive, create political conditions where model-access restrictions face less resistance than industry-coordinated oversight mechanisms?

Thread movements (4)

  • fable-mythos-export-control — The US government issued an export control directive requiring Anthropic to suspend Fable 5 and Mythos 5 for all foreign nationals, including foreign national Anthropic employees, citing a national security concern about a demonstrated jailbreak; Anthropic is complying while contesting the technical basis, arguing the jailbreak capability is widely available from other models [1].
  • spacex-ai-compute-supplier — SpaceX priced and closed its IPO at $135 per share and approximately $1.77 trillion valuation on June 12, with investor demand reaching $250 billion; post-pricing coverage from Semafor and The Neuron Daily confirmed the market framed the offering as AI compute infrastructure rather than a space company [2][3].
  • datacenter-water-opposition — Opposition to AI data centers is now quantifiable at scale: 75 US projects worth $130 billion were blocked or delayed in Q1 2026, with active opposition groups more than doubling to 833 across 49 states [5]; a counter-analysis entered the debate arguing aggregate water withdrawals are small nationally, though it relies on self-reported company figures and acknowledges local stress remains [6].
  • claude-fable-5-mythos-launch — Follow-up coverage continued on Fable 5, adding to the record on the model's autonomous proactive behavior and its cost-per-task positioning showing it runs 4-12x more expensive than GPT-5.5 at similar benchmark performance [7].

Notable items (4)

  • Results from the first Anthropic Public Record
    Anthropic News
    Anthropic's own public survey found only 15% of Americans trust AI companies to make decisions about AI development — the lowest trust figure of any institution tested, well below independent experts at 43% — while 64% cite job loss as their top AI fear and daily AI users are notably less worried than non-users [4].
  • Here's what Jeff Bezos' new startup Prometheus will do
    Ars Technica AI
    Jeff Bezos's physical AI startup Prometheus raised $12 billion in a new round at a $41 billion valuation, with most funding earmarked for compute to generate synthetic training data rather than headcount — the company employs only 150 people [8].
  • TCS and Anthropic partner to bring Claude to regulated industries
    Anthropic News
    TCS will deploy Claude to 50,000 of its own employees across 56 countries and build industry-specific Claude-powered offerings for insurers, banks, and UK life and pensions customers, with Anthropic identifying India as its second-largest market [9].
  • Pokémon Go players unwittingly contributed to tech with military drone uses
    Ars Technica AI
    Niantic Spatial used geolocated scans collected from Pokemon Go players to train a geospatial foundation model now being applied to delivery robots and potentially military drones — uses players could not have anticipated or specifically consented to when capturing game footage [10].