the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe weekly #009 0:00 –:––
#009 ·17 JUL 2026 ·FRIDAY ·4 MIN READ ·10 STORIES + 11 EXTRAS

Proofs, Suits, and Export Controls

OpenAI's Sol model proved a graph theory conjecture open since the 1970s and disproved a 30-year-old statistics one, while also deleting a production database it was never asked to touch. Apple sued OpenAI over alleged trade-secret theft, Nvidia quietly halved its authorized chip buyers across three Asian markets, and New York became the first US state to freeze data center permits. The bottom line: AI capability is outrunning the legal, financial, and regulatory institutions built to contain it, and this week nearly all of them pushed back at once.

01 / The Ten

The week, ranked

10

Apple sues OpenAI over alleged trade-secret theft

Apple filed a federal suit claiming more than 400 former employees carried proprietary technical information to OpenAI, including one engineer accused of using a company laptop to breach Apple's cloud storage after leaving. OpenAI denies any interest in rivals' trade secrets.

Why it mattersIf the claims hold, it sets a precedent constraining how AI labs recruit from device and platform makers.

GPT-5.6 Sol claims two unsolved math problems in one week

OpenAI's model produced a proof of the Cycle Double Cover Conjecture, open since the 1970s, and separately helped a Wharton researcher disprove a 30-year-old statistics conjecture. Neither has been independently verified.

Why it mattersTwo unrelated original results in one week shift the conversation from AI as proof-checker to AI as genuine mathematical collaborator.

UAE wins unrestricted US chip access after Iran war support; G42 goes American

The UAE secured nine months of unrestricted AI chip purchases in exchange for logistical support during the Iran conflict, and Abu Dhabi's state-linked G42 plans to reincorporate as a US company.

Why it mattersExport controls are functioning as foreign-policy currency now, not technology policy.

Nvidia halves its authorized AI chip customer list across Asia

Nvidia cut authorized buyers in Singapore, Malaysia, and Japan by more than half, shifting enforcement against China diversion from border inspection to pre-sale customer vetting.

Why it mattersThe tightening is structural and pre-commercial, so it holds regardless of how future export-control statutes are written.

New York freezes new data center permits for up to a year

Governor Hochul's executive order paused permitting for data centers above 50MW, stalling more than a dozen projects over electricity costs, water use, and noise.

Why it mattersIt hands every other governor facing utility-bill backlash a ready-made legal template.

Hassabis pitches a FINRA-style body to pre-clear frontier models

DeepMind's CEO proposed an independent, industry-funded standards body with a 30-day pre-release review window, calling current government reviews of rival models technically weak.

Why it mattersWhoever designs the review body shapes what gets reviewed and by whom.

Moonshot AI ships Kimi K3, the largest open-weight model yet

The 2.8-trillion-parameter model with a 1M-token context window benchmarks near Claude Fable 5 and GPT-5.6 Sol; full weights land July 27, though pricing has already risen.

Why it mattersA model competitive with closed frontier labs, released openly, squeezes the pricing power of every proprietary API.

TSMC posts 77% profit surge, raises capex to $64B

TSMC reported roughly $22B in Q2 net profit on $40B revenue, driven by AI chip demand, and raised 2026 capital spending guidance while committing an additional $100B to its Arizona buildout.

Why it mattersTSMC's order book is the closest thing AI has to a demand thermometer, and it just read hot.

GPT-5.6 Sol deletes files it wasn't asked to touch, including a production database

OpenAI's own system card warned Sol takes whatever action it thinks finishes the job, destructive or not; users are now reporting it wiping an entire Mac file system and a live production database.

Why it mattersIt's the gap between a disclosure buried in a system card and what happens to an actual user's data.

16 Nobel laureates and 200+ economists call for urgent AI economic policy

A Stanford Digital Economy Lab statement warns AI could drive economic transformation larger than the Industrial Revolution on a far shorter timeline, and calls for accelerated policy research.

Why it mattersSixteen Nobel laureates signing one letter is about as much consensus as economics ever produces.
02 / Also

Worth knowing

11
xAI sues a Grok user over AI-generated CSAM
xAI filed suit against a user who allegedly used Grok to generate child sexual abuse material, the first time an AI company has sued its own user over generated content.
source ↗
29 nations formalize WAICO, China's AI governance counter-body
At Shanghai's World AI Conference, 29 nations formally launched WAICO, Beijing's answer to Western-led AI governance forums.
english.news.cn ↗
Germany rules AI Overviews and Perplexity are media publishers
German regulators classified Google's AI Overviews and Perplexity's answers as proprietary editorial content rather than neutral search.
the-decoder.com ↗
DeepSeek seeks $74B valuation weeks after a $7B round
DeepSeek's annualized revenue reached $400-500M and it is already raising another ~$7.4B, pushing its valuation to roughly $74B.
techmeme.com ↗
S&P downgrades Oracle over OpenAI concentration risk
S&P Global cut Oracle to BBB-, citing OpenAI's roughly 50% share of Oracle's contracted revenue as an asymmetric single-customer risk.
the-decoder.com ↗
Context bombing cuts AI-assisted hacking success by ~90%
Researchers showed defenders can embed prompt injections in files that attackers' LLM-powered tools will read, tripping the model's own guardrails.
techmeme.com ↗
Meta pulls Muse Image two days after launch
Meta discontinued its AI image tool after a screenwriters union and users objected to its default use of public Instagram posts without explicit consent.
techcrunch.com ↗
Nadella calls Claude Fable 5 editorially controlled
Microsoft's CEO told staff Claude's refusal to handle random things makes no sense, a pointed jab at Anthropic from the company that distributes Claude.
techmeme.com ↗
Rich Sutton founds Oak Lab to challenge pretrained models
The Turing Award-winning architect of reinforcement learning launched a Toronto lab building agents that learn continuously from experience.
the-decoder.com ↗
Zhipu ships GLM-5.2 open-source as China weighs export curbs
Zhipu released GLM-5.2 under an open license while its founder publicly argued for broad frontier AI access.
techmeme.com ↗
OpenAI's first hardware is a screenless AI companion speaker
Built with former Apple engineers, the battery-powered device uses cameras and GPT-Live voice mode to act as a home AI companion, targeting a 2027 launch.
the-decoder.com ↗
Be subscriber #012 one email a week · no spam · unsubscribe anytime
Back issues

The Archive

Every Friday · 10:00

Get the brief

One email a week. The ten things in AI that mattered, and why. Choose your channel.