What mattered · week 46
Nvidia's CEO is at the centre of a US-China spy scandal; METR finds AI is roughly doubling the speed of AI progress at Anthropic; and Claude Fable 5.2 nearly doubles the best score on a remote-work benchmark.
Espionage allegations reach the heart of the chip race, an outside evaluator finds AI is speeding up AI research, and Anthropic ships a new top model.
What mattered
Nvidia's CEO is at the centre of a US-China spy scandal. Jensen Huang admits an affair with a woman US intelligence reportedly believes worked for Beijing. You couldn't make this up. The Information reports, though Reuters couldn't verify it, that the relationship helped move at least $7.5 billion of restricted chips into China. Nvidia lost about $190 billion in a day, its board is investigating, and the Trump administration has questions to answer.
No one has been charged. But if the reports hold up, they show how badly these chips are wanted, and how long a game China is willing to play. Read more
METR finds AI is roughly doubling the speed of AI progress at Anthropic. The independent evaluator estimates Anthropic's models have improved at about twice their earlier rate since October 2025 (plausible range 1.3× to 3×). It guesses about 80% of that comes from Anthropic's use of AI, and saw no bottleneck to stop it compounding over the next year. Anthropic redacted most of the key evidence. Read more
Anthropic's Fable 5.2 nearly doubled the best score on a remote-work benchmark. Released on 9 November, ten weeks after Fable 5.1, it makes its biggest gains on long-running agentic tasks. Anthropic still recommends Opus 5.5 for everyday work.
| Remote Labor Index | Score |
|---|---|
| Fable 5.2 | 41.3% |
| Opus 5.5 | 21.3% |
| GPT-6 Astra | 20.8% |
| Fable 5.1 | 17.9% |
Vendor figures (Anthropic), 9 November 2026.
One thing to try
Not sure which AI to pay for? Start with my two-minute guide. It compares the two leading choices for everyday knowledge work, Claude and ChatGPT, both about $20 a month. Releases like Fable 5.2 change which model is strongest month to month; the advice is about the app, so it holds.
First step: open the guide, pick one, and choose the $20 plan at claude.ai or chatgpt.com. Five minutes. Which AI should I use?
The bigger picture
AI 2027 keeps reading less like fiction. The scenario turns on labs using AI to speed up their own research, and on espionage over the US lead. This week brought evidence of the first and allegations of the second. Read more
What did I miss? Just reply to this email. The rest of what I log is at wayintoai.com/logs.