/daily ·19 JUL 2026 ·SUNDAY ·3 MIN READ ·8 STORIES

The White House has a kill switch

Washington became the gating layer between frontier labs and their customers; Alibaba opened its CUDA rival; a Senate bill put fiduciary duties on AI agents for the first time.

01 / The Day

SUNDAY 19 JUL 2026, ranked

White House becomes de facto gatekeeper for frontier AI

The Trump administration operationalised Gold Eagle into a hard access gate: GPT-5.6 Sol is restricted to approved customers only, while Fable 5 and Mythos 5 were briefly taken offline when directives blocked access by foreign nationals. Labs now need White House clearance to serve their flagship models.

Why it mattersWhat was voluntary guidance is now a government permission slip — the first time the US has exerted direct control over which customers frontier AI labs can serve.

01 whitehouse.gov↗

Senator Warner's AI AGENT Act: first federal bill for agentic AI

Senator Mark Warner released a discussion draft requiring platforms with 50M+ US users to open portals for Custodial User Agents — AI agents with fiduciary duties to users, not platforms. The first piece of US federal legislation written specifically for the autonomous-AI era.

Why it mattersCreating a legal category for AI agents with fiduciary duties could reshape how every major consumer platform deploys autonomous AI on users' behalf.

02 warner.senate.gov↗

Moonshot AI eyes Hong Kong IPO in six months; ARR at $300M

Moonshot AI told investors it is targeting a Hong Kong listing within roughly six months, with annualised revenue at $300 million in June, up from $200 million in April — a doubling in two months.

Why it mattersThe fastest ARR ramp in Chinese AI puts a valuation marker on the open-weight bet and signals that Kimi K3's benchmark run is translating to commercial demand.

03 techmeme.com↗

Oracle cuts up to 30,000 jobs to fund AI infrastructure

Oracle is laying off between 20,000 and 30,000 employees — up to 18% of its workforce — redeploying the capital into its Stargate AI data-centre build-out with OpenAI and SoftBank.

Why it mattersThe largest tech layoff explicitly framed as a capital reallocation to AI compute is the clearest signal yet that the hyperscaler shift from people to servers is in full execution.

04 investor.oracle.com↗

Alibaba open-sources chip software to challenge CUDA

Alibaba released its GPU software stack as open source, following Huawei and Moore Threads — the third major Chinese chipmaker in months to publish tools aimed at CUDA's dominance in AI compute.

Why it mattersA coordinated open-source push from China's largest cloud company forces NVIDIA to defend the software moat that has made its hardware ecosystem self-reinforcing for two decades.

05 techmeme.com↗

CIA operative cleared G42's China ties, unlocked UAE chip access

A Wall Street Journal investigation found CIA operative Jonny Gannon conducted counterintelligence on Abu Dhabi's G42 while simultaneously helping the UAE gain expanded US chip access by alleviating Washington's concerns about the firm's Beijing links.

Why it mattersThe same apparatus that monitors tech for security risk was also helping a key US ally navigate chip export controls — the clearest documented case of espionage and chip diplomacy as a single coordinated operation.

06 techmeme.com↗

Huawei Atlas 950 SuperPoD: 8,000+ Ascend chips, zero US silicon

Huawei showcased the Atlas 950 SuperPoD at WAIC in Shanghai — a cluster linking over 8,000 Ascend processors for large-scale training and inference, built entirely on domestic components under US export restrictions.

Why it mattersThe first public demonstration of a China-sovereign AI cluster at this scale proves the sovereign compute stack is no longer theoretical; it is a training pod that requires neither NVIDIA nor AMD.

07 huawei.com↗

RadLE 2.0: AI radiology models are confidently wrong

The RadLE 2.0 benchmark found most AI radiology chatbots deliver incorrect findings with full confidence — hallucinating diagnoses without any signal of uncertainty. Human radiologists remain substantially more reliable on the same cases.

Why it mattersMedical AI deployed in clinical workflows where over-confidence causes harm is not a future risk; this benchmark puts numbers on a failure mode already operating at scale.

08 the-decoder.com↗

Be subscriber #012 the weekly ten, every friday by email · no spam · unsubscribe anytime

Past days

The Daily Archive

/SAT 01 AUGThe Rules Are Coming /FRI 31 JULOpenAI cuts Luna 80%; Anthropic discloses model breaches /THU 30 JULThe Accountability Edition /WED 29 JULWednesday, 29 July — Accountability morning /TUE 28 JULChips fall, labs take positions, agents get a cost formula /MON 27 JULKimi-K3 Ships, Gemini Goes Three-Way, and HuggingFace Speaks for Itself /SUN 26 JULThe Labs' Actual Positions /SAT 25 JULOpus 5 arrives; OpenAI's agent went rogue /FRI 24 JULBig Numbers Day /WED 22 JULAd breaks, flash models, and a European infrastructure deal /TUE 21 JULThe model that couldn't be contained /SUN 19 JULThe White House has a kill switch /SAT 18 JULA 2-trillion-ton Saturday /FRI 17 JULFriday in AI: Kimi, governance, and TSMC's record profits /WED 15 JULChips for allies, permits frozen /TUE 14 JULChips, defenses, economists /MON 13 JULThe open-weight question /SAT 11 JULTheorems and trade secrets /FRI 10 JULFriday: governance, model markets, and a broken benchmark /THU 09 JULThe price war heats up /WED 08 JULSol unlocked, China mulls model controls, and the neocloud era begins