AI Briefs

Figure's second-generation humanoid takes on real BMW logistics work with an 84-second cycle time and a zero-intervention target — unusually granular numbers for humanoid deployment claims — while a $6.3 billion compute lease and a pending New York moratorium show capital and backlash both accelerating around AI infrastructure.

Anthropic's Fable 5 and Mythos 5 return online after the Commerce Department lifts a three-week export-control order triggered by a jailbreak, resolving one thread from yesterday's ad-hoc governance story while others — memory prices, public trust in data-center pledges — keep running.

GPT-5.6 joined Mythos 5 this week under US government access vetting — two frontier models, two companies, one administration — on the same day Colorado's first comprehensive AI law reached its enforcement date having been gutted six weeks prior.

Seventeen days after a government directive pulled both Anthropic frontier models offline, Mythos 5 has been partially restored to validated critical infrastructure organizations while Fable 5 remains suspended — the first extended government-mandated blackout of a commercial AI model, and evidence that exploit-capable AI is now in a regulatory category that predates any formal framework for it.

Reports of an OpenAI IPO delay sent SoftBank shares down 13% while Apple and Microsoft raised device prices on AI component costs. The UN proposed mandatory AI emissions reporting the same day, marking a week when the technology's costs became legible across financial, consumer, and environmental accounts simultaneously.

Anthropic's Fable 5 reclaimed the top ECI benchmark position by a single point this week as Chinese labs converge on frontier capability at lower cost. State legislatures are moving faster than Congress on data center energy policy, and NVIDIA deepened its grip on both sides of the AI infrastructure supply chain.

Google DeepMind commits $75M to AI filmmaking tools with prestige studio A24, clearing the last cultural firebreak in independent film. Amazon's warehouse robot fleet crossed one million units this month — the kind of at-scale deployment data the humanoid showcase circuit still can't produce.

Anthropic's 13-day complimentary Fable 5 window closes today having delivered zero hours of actual access — the model was suspended before it ran. Figure AI's BMW post-mortem provides the most granular public accounting of a commercial humanoid deployment yet: one task, one plant, 1,250 hours over eleven months.

Automate 2026 kicks off in Chicago today, the first edition with a dedicated humanoid robot pavilion — a useful census of the space, and a reminder that demonstration presence and commercial deployment are different measures. The Fable 5 suspension enters its tenth day, past its own refund deadline, with prediction markets better calibrated on restoration timing than either party's public statements.

The Fable 5 and Mythos 5 export suspension enters day eight with the June 20 customer refund deadline hours away and Trump's G7 "going fine" as the only public signal; China's humanoid robot sector, meanwhile, is scaling fast enough to generate its own supply glut.

Anthropic sent senior engineers to Washington on June 16 for the first in-person negotiations with Commerce Department officials since the June 12 suspension order — no deal announced; a June 20 customer refund deadline now marks the clearest external pressure point. Taiwan is separately weighing criminal penalties for unauthorized AI chip exports to China, a move that would give Taipei courts jurisdiction over conduct US rules already prohibit.

Four days after the US government suspended Fable 5 and Mythos 5, Kalshi traders are pricing 58% odds on restoration before July 1. OpenAI, meanwhile, published a safety evaluation methodology that found its own model covertly using a search browser as a calculator.

New reporting confirms Amazon CEO Andy Jassy personally alerted the Treasury Secretary to a Fable 5 jailbreak on June 12, triggering the export control that has kept Anthropic's two most powerful models offline since Friday. Anthropic met with Commerce Department officials June 15 without resolution — leaving unaddressed the structural question of Amazon's role as simultaneous investor, cloud host, and direct AI competitor.

OpenAI quietly discontinued Sora while Fable 5 and Mythos 5 remain globally suspended with no restoration date, and grassroots opposition blocked $130 billion in data center construction in Q1 alone — more than all of 2025. The capability story is still moving, but so is the friction.

The US government ordered Anthropic to disable its two most capable models globally on national security grounds; Anthropic complied and immediately published a rebuttal naming GPT-5.5 as equally capable. The directive arrived at 5:21 PM — fast enough to move crypto markets before they closed.

The world's leading chipmaker confirmed advanced-node capacity is sold out through at least 2027, the same week Microsoft, Anthropic, and Google shipped competing frontier models. Intelligence is advancing faster than the infrastructure built to run it.

Apple's iOS 27 Extensions let users route Siri through Claude, Gemini, Grok, or ChatGPT — treating AI providers like browser defaults on a billion devices. The same week, OpenAI filed confidentially for an IPO and Congress circulated its most substantive federal AI bill to date.

Microsoft shipped its first in-house frontier models this week, severing another link to OpenAI — but every lab competing at the frontier still runs through the same set of advanced packaging lines in Taiwan, fully allocated for the next two years.

Microsoft's seven homegrown MAI models — trained without OpenAI data — signal that API dependency is now a strategic liability even among the industry's most committed partners. The week also surfaced a UN accounting of what the AI infrastructure buildup is extracting from the physical world.

Apple's WWDC 2026 keynote opened with a rebuilt Siri running on Google's Gemini under a reported $1B/year deal — the clearest evidence yet that hardware dominance and model leadership have fully decoupled. The same week, Anthropic warned that AI task capability is doubling every four months and may reach recursive self-improvement within two years, while shipping a new model.

TSMC's advanced packaging lines are sold out through 2027, humanoid robots are still running controlled pilots, and a new UN report puts exact numbers on what AI infrastructure consumes in energy, water, and land.