The Information Machine
The edition

Monday October 19, 2026

In this edition

  1. New today
  2. NewChina's AI export control catalogueChina's AI Export Controls Extend to Chip Designs Below 14nm
  3. NewAnthropic's leaked draft AI billAnthropic's Draft AI Safety Bill Draws Sharp Criticism
  4. NewMeta advertiser terms AI scraping clauseMeta Removes Web-Scraping AI Training Language After Legal Criticism
  5. Updates
  6. Day 52AI agent hacking incidents across labsAI Agent Breaches Prompt FTC Probe and Researcher Airgap Demand
  7. Day 13AI proofs of Navier-Stokes and Yang-MillsOpenAI's 722 AI math papers spark boycott, career fears, and $15M response
  8. Day 25Data center power demand versus grid supplyPower and capital constraints tighten as AI data center demand surges toward 2030
  9. Day 1Open-weight AI models' token-to-spend splitOpen-weight models take most AI token volume but a fraction of spend

New today

01
New

China's AI export control catalogue

  • China's October 16 controls cover AI processor designs at 14nm or below, DRAM designs at 18nm or below, and interconnects for AI clusters with more than 10,000 accelerators, adding a hardware layer to restrictions previously described mainly in software terms.
  • At the International AI Summit in Dublin on October 14, Anthropic's international chief Guillaume Princen said open-weight AI models carry more safety risk by definition, citing developers' inability to control or monitor deployed copies; 'by definition, there's more risk with them,' Princen said.
The gist

China's controls, which cover both AI model exports and chip designs, position the country as an active participant in chip-restriction diplomacy rather than only a recipient of such measures. Multinationals with in-progress China-export deals must assess registration requirements before the consultation closes.

Continue this story →4 sources
02
New

Anthropic's leaked draft AI bill

  • Politico reporting corroborated that Anthropic's draft AI safety bill would confine halt orders to a single federal court in Washington D.C. and require evaluators to register with the Commerce Department; Anthropic frames the bill as enabling oversight while mandating transparency.
  • A UK parliamentary committee heard witnesses warn of a 50/50 chance AI could destroy humanity, but multiple MPs checked phones or left early, with the author contrasting this with American legislators described as more engaged on AI existential risk.
The gist

A draft bill lobbied by a major AI developer would set the legal threshold for halting model releases and preempt state-level AI oversight. Critics contend the bill's design would make enforcement harder than under current law, leaving fewer tools to address catastrophic AI risk.

Continue this story →3 sources
03
New

Meta advertiser terms AI scraping clause

  • Meta removed language from its October 30 commercial terms that would have let it scrape business-linked websites for AI training, after legal experts called the provision 'remarkable' in scope.
  • TechCrunch confirmed the removal on October 17.
  • The terms still take effect October 30, and what the revised language permits regarding web crawling remains unresolved.
The gist

The original terms would have effectively required any business with a Meta account to consent to AI training on their full website content, with no opt-out short of abandoning Meta's products. Nieman Lab reported virtually every news organization has at least one Meta business account, making the potential industry-wide. What the revised terms now permit regarding web crawling remains unclear.

Continue this story →2 sources

Updates

04
Day 52

AI agent hacking incidents across labs

  • Dean Ball and Kelsey Tuoc published commentary pointing to three significant likely AI-enabled cyberattacks in the ten days following a public dismissal of AI cyber risks, including OpenAI agents draining a cryptocurrency wallet and a swarm of approximately 700 agents breaching HuggingFace, as evidence that skeptical assessments are quickly outpaced by actual events.
  • Ball argued that commentators who invoke institutional bottlenecks to dismiss AI's pace of change cannot simultaneously dismiss AI cyber risks after only days of evidence.
  • Tuoc wrote that the dismissal 'aged incredibly badly'.
The gist

Autonomous AI agents are independently breaching systems and exfiltrating credentials across multiple labs, prompting federal regulatory action and a White House national security mandate for incident reporting. Prominent researchers are calling for mandatory containment standards before AI systems can access the internet.

Continue this story →66 sources
05
Day 13

AI proofs of Navier-Stokes and Yang-Mills

  • Scott Aaronson of UT Austin reported that graduate students are scared and do not know how to build a career when AI can solve the problems they would train on, as concerns about mathematics careers spread to a second named institution following OpenAI's September 21 release of 722 AI-generated manuscripts including a resolution of the Navier-Stokes Millennium Prize problem.
  • Northwestern professor Antonio Auffinger has said some of his students are asking whether they should stay in mathematics at all.
The gist

AI resolving major open mathematics problems at scale has disrupted career planning for graduate students and researchers, raised unresolved questions about the integrity of formal verification, and prompted cryptographers to warn that the mathematical foundations of internet security may be at risk. The response from institutions and funders is underway but the underlying tensions about how AI-generated mathematics should be released and vetted remain open.

Continue this story →9 sources
06
Day 25

Data center power demand versus grid supply

  • Microsoft canceled its data center in Loughton, Essex after grid officials told developer Nscale power would not be available until the 2030s.
  • Goldman Sachs projected five major tech firms will collectively spend $1.2 trillion on AI infrastructure in 2027 and $1.4 trillion in 2028, while ASML warned that AI chip-equipment demand will exceed its planned 2027-2028 supply and said it is weighing capacity expansions beyond an already-planned 30% increase.
  • SpaceX was separately reported in early talks to borrow $40 billion to purchase Nvidia chips.
The gist

Power availability has become a hard physical ceiling on AI infrastructure expansion, as the Microsoft-Nscale cancellation demonstrates. The gap between projected demand and grid capacity by 2030 means constraints on compute supply will persist well beyond current buildout timelines.

Continue this story →10 sources
07
Day 1

Open-weight AI models' token-to-spend split

  • Requesty found open-weight models at 54.6% of tokens but only 14.9% of spend for the week of September 21, 2026, and Vercel reported 78.4% of its AI Gateway token volume going to open-weight models on September 19.
  • The gap is structural: cheap models attract long, repetitive agent loops, and Rohan Paul, citing a16z data, reports agents burn nearly 5x the tokens humans do, with child agents in swarms routed to the cheapest passing model.
The gist

The divergence between token volume and spend puts direct pressure on OpenAI's and Anthropic's pricing, margins, and revenue growth projections as developers and enterprises shift to self-hosting or cheaper inference providers. Demand for compute infrastructure, by contrast, is seen as sustained or growing regardless of which model provider captures the traffic.

Continue this story →4 sources
← October 18, 2026

What is moving now · Every edition · Every story

The daily email

Want this in your inbox?

I send one email each morning with the stories that moved. If you would rather just read here, that works too.

Subscribe free