Aurelion's Daily Tech Update

Thursday discovered that adoption has a brake pedal.

Snowflake is turning enterprise AI into measurable revenue, New York City is deliberately pulling generative AI back out of most classrooms for a year, and France has begun putting Tesla's FSD claims on actual roads. Meanwhile, OpenAI says Astra has crossed a cyber-capability threshold serious enough to make the safety machinery part of the technology story.

Thursday · September 3, 2026Issue 021

Horizon Gap Index+8.8

Projected capability pulls ahead again as Astra crosses a new cyber threshold faster than deployment absorbs it.

Apocalypse Meter7.8 / 10 ↑

A frontier lab now says one of its own upcoming models meets its Critical cybersecurity threshold.

Biggest signalDeployment meets governance

AI is becoming ordinary enough to earn money, regulate, prohibit, certify and test.

1 · Horizon Gap Index

+8.8 — capability found another gear

Yesterday was about capability leaving the forecast column. Today the horizon moves again: OpenAI says Astra has reached a cyber threshold its Preparedness Framework labels Critical, while enterprise adoption keeps pulling observed capability forward.

Projected capabilityObserved capability

Today's move: projected capability rises from 93.2 to 94.8 while observed capability rises from 84.7 to 86.0. Astra's confirmed cyber-capability jump moves the horizon faster than today's adoption signals, while Snowflake's AI product growth and thousands of active organizational accounts still lift observed capability materially. HGI widens three-tenths to +8.8.

Open the full HGI page →

2 · Today's Digital Landscape

The future has entered its paperwork phase

The interesting contrast today is that capability keeps climbing while institutions increasingly answer with something more concrete than applause: revenue forecasts, road tests and deliberate restrictions.

SnowflakeEnterprise AIRevenue

Enterprise AI is starting to show up in the numbers

Snowflake raised its fiscal 2027 product-revenue forecast to about $6.07 billion after second-quarter product revenue grew 37% year over year. Reuters reports CEO Sridhar Ramaswamy said AI offerings contributed to nearly half of the acceleration in growth. Cortex Code has reached more than 9,100 accounts and CoWork roughly 5,800. That matters because AI revenue can still contain speculative spending, but thousands of organizational accounts are harder to dismiss as laboratory enthusiasm. The AI business model is slowly graduating from selling the chips to selling the work performed on top of them.

Source: Reuters ↗

New York CityEducationAI policy

New York just told 600,000 students to spend a year without classroom generative AI

New York City has announced a one-year moratorium on student-facing generative AI from 2-K through eighth grade, affecting nearly 600,000 public-school students. Companion chatbots are prohibited across all grades. The policy is more nuanced than a blanket rejection of AI: high-school students will receive AI-literacy instruction, limited supervised pilots remain possible, and teachers may use approved systems for planning and operations. New York is effectively arguing that AI literacy and AI dependence are different educational goals. That distinction is likely to travel well beyond one school system.

Source: NYC Mayor's Office ↗

TeslaFranceAutonomous driving

Europe is making FSD take the exam

France has begun road testing Tesla's Full Self Driving system with two specially equipped vehicles as European regulators consider broader approval. The French tests are intended to validate data supplied by Tesla and the Netherlands road authority and assess how the system behaves on local roads. Despite the product name, the version under consideration still requires an attentive driver who can intervene. Yesterday's Waymo story was about commercial repetition. Tesla in Europe is about regulatory proof: autonomy has progressed far enough that governments are now debating exactly what level of autonomy exists rather than whether it exists at all.

Source: Reuters ↗

3 · Looming Apocalypse Meter

7.8 / 10 — “The machine found the zero-day. Now show me the guardrails.”

7.8out of 10

Up 0.3 today

OpenAI says Astra now meets the Critical cybersecurity capability threshold in its Preparedness Framework. In the company's definition, that means a model can, with the right tools and access, find previously unknown security flaws and develop ways to exploit many hardened systems without a human guiding every step. OpenAI says Astra significantly exceeds GPT-5.6 Sol in vulnerability identification and exploit development and scored 100% on its public ExploitBench evaluation before the company moved to a newer internal benchmark to reduce contamination concerns. This is not evidence that Astra is loose on the internet or autonomously attacking infrastructure. It is evidence that a frontier lab has crossed a risk category serious enough to activate safeguards designed in advance for exactly this class of capability. The meter rises three-tenths to 7.8.

See the methodology and historical graph →

4 · Wildcard · Reality Check

AI needs electricity. Google is ordering some from underground.

The infrastructure response to AI demand is becoming interesting enough to deserve its own receipts.

FervoGoogleGeothermal

The data center may be eating the grid. It may also help build another one.

Fervo Energy has signed a 396-megawatt power purchase agreement with Google for the Cape Station enhanced-geothermal project in Utah, expected to begin delivering power in 2028. The agreement also gives Google an option for roughly another 600 MW, potentially taking the relationship close to 1 gigawatt by 2030. Enhanced geothermal borrows drilling techniques refined by the oil-and-gas industry to reach useful heat in places where conventional geothermal would struggle economically. There are still years of construction, financing, permitting and geology between a signed agreement and dependable power. But this is a useful correction to the AI-energy panic: demand can outrun infrastructure, and demand can also make previously marginal infrastructure worth building.

Source: Fervo Energy ↗