Good morning! Today's big story is a new top model. Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, and Fable now leads the Artificial Analysis ranking. It lands a few days after METR said Anthropic's models improve about twice as fast as a year ago. It also lands in the week Anthropic is expected to set its IPO price range.

Look at the rest of the news and one word keeps coming back: trust. Nvidia's board is investigating its CEO over ties to a suspected Chinese spy, and the warning signs were there a year ago. Coding agents keep touching our keys and passwords. Gemini faces a new claim about human reviewers, and users ask if Opus 5.5 was quietly changed. Each time, the public learns late, and the only ones who can really check are the companies themselves.

That is my worry with this speed: the models change every few weeks, the rules and the checks move much slower. Money from a stock market sale will not slow anything down. Australia's push for incident reporting with a deadline is the kind of answer we need more of. To watch: Anthropic's valuation range, Nvidia's board, and whether Google and Anthropic answer this week's claims.

Today's Morning Tokens:

1. Anthropic launches Claude Fable 5.2 and Mythos 5.2, days before setting its IPO price
2. Nvidia scandal: Huang admits affair, stock slides, lawyers circle
3. Your coding agents and your keys: Claude Code secrets hit the news, ChatGPT Desktop questioned
4. New claim: Gemini sends personal Drive files and emails to human reviewers
5. Was Opus 5.5 quietly weakened on November 3? Users ask, nobody can check
6. Australia moves toward deadlines for reporting AI incidents
7. Meta's Muse Ember: a cheap small model for the work agents repeat all day

Read it all, with the sources: https://aidailygrind.com/2026-11-10.html

Read by an AI voice. Stories are picked and written with the help of AI tools, from public news sources. Always check the sources.

A fee when machines replace workers. A good idea. But who checks, and who cheats?

Robot tax: should companies pay when AI takes your job? And if so, who makes sure they really pay?

In this Slow Brew, I explain the robot tax, also called an automation levy, in plain words. It has three main versions: a fee per job replaced, a tax on AI tokens or AI revenue, and a tax on data centers.

We look at where the idea has been: Bill Gates in 2017, the European Parliament saying no that same year, and South Korea cutting a tax break for automation. Then we look at today: the Sanders Senate report, a House bill that taxes AI tokens, and the data center ideas from Senators Warren and Wyden. We also cover OpenAI's own April 2026 paper asking for taxes on automated labor, and the San Mateo County rule in California that took effect on October 29.

Why it matters: our pensions and health care are paid by taxes on work. If AI does more of the work, that money must come from somewhere. But how do you check a company? And what stops work from moving to countries with no tax?

I tell you honestly that I have not made up my mind on this one, and I want to hear where you stand.

Follow the AI Daily Grind podcast so you don't miss the next Slow Brew.

If this was useful, follow the show and save the episode: it helps other people find it.

Read it, with every source: https://aidailygrind.com/deeper-robot-tax-idea.html

Read by an AI voice. Stories are picked and written with the help of AI tools, from public news sources. Always check the sources.

Bonjour, and happy Monday. Today's news has one question running through it: who is checking? Nvidia's board opened an independent investigation into its own CEO, Jensen Huang, after reports that China recruited a woman to spy on him. Goldman Sachs is pulling Claude Code from its bankers and traders, and our sources do not say why. A Reddit post also shows that Google quietly changed a Gemini privacy promise on October 16.

See the pattern? AI goes deeper into our chips, our banks, our inboxes and our terminals, and the checks arrive late. A promise edited in October gets noticed in November. A second user says a local Qwen agent fetched web addresses nobody asked for. October was full of incidents like this, and November is not slowing down. If you use agents at work, this is the week to check what they can touch, and to ask your vendors how fast they will tell you about changes.

There is good news too: big open models now run on cheap home hardware, a real win for privacy. To watch: the Nvidia investigation, a reason from Goldman, a reply from Google, and the US Congress, back this week, where rules for the most powerful AI are on the table. Rules are welcome, voilà, but they must be checked by independent auditors, not written by the labs alone.

Today's Morning Tokens:

1. Nvidia's board investigates CEO Jensen Huang over ties to an alleged Chinese spy
2. Goldman Sachs pulls Claude Code from its bankers and traders
3. Google quietly changed a Gemini promise: human reviewers may see your connected files
4. A second user says a local Qwen agent fetched web addresses nobody asked for
5. Self-improving agents learn new tricks, and forget old ones
6. Big AI on small machines: a 284B model with no GPU
7. Elvis and Eminem rights holders sue Suno

Read it all, with the sources: https://aidailygrind.com/2026-11-09.html

Read by an AI voice. Stories are picked and written with the help of AI tools, from public news sources. Always check the sources.