CW

[CW @willywill8604]

On July 15, hundreds of millions of people in China woke up to a notification: your AI companion has been deleted.

ByteDance, Alibaba, and Tencent shut down personalized AI agents overnight. Not because of a flaw. Because the features that make companions feel personal are the same features that make them impossible to regulate.

The U.S. has no equivalent regulation. https://t.co/wcpsHGuDKP and Replika operate without any of the safeguards China just mandated.

Is China ahead on AI safety, or is this censorship wearing a safety hat?

https://t.co/LwBjCiI1dM

Next one drops July 26.

[CW @willywill8604]

@shairnn 18

[CW @willywill8604]

@michelleerrq now show the clip fixing all the answers that were marked incorrect 🤭

[CW @willywill8604]

GPT-5.6 was voluntarily gated for 13 days. Released with government blessing.

Fable 5 was forcibly banned for 19 days. Restored with ID checks and billing restrictions.

Both models have comparable cyber capabilities. Both triggered by the same executive order.

If the government treats cooperators and resisters differently — is the "voluntary" framework actually voluntary?

Thirteen weeks of the Signal Brief. Every Sunday. Quarter one complete.

https://t.co/0aEER4lrBG

[CW @willywill8604]

Fable 5 came back on July 1 after 19 days offline.

During those 19 days, China's open-weight GLM-5.2 was independently verified to match Mythos on the exact capability that triggered the ban.

The capability the export controls were designed to contain is now freely downloadable by anyone on earth.

Did the ban protect anything, or did it prove that export controls can't outrun open-source?

https://t.co/E21dMUDDpE

Next one drops July 12.

[CW @willywill8604]

The co-author of "Attention Is All You Need" — the paper that started the entire AI era — just left Google for OpenAI.

Google paid $2.7 billion to bring him back 22 months ago.

If the person who invented the architecture that made all of this possible keeps leaving Google — what does Google's research culture look like from the inside?

That, plus ChatGPT dropping below 50% market share for the first time and the Trump AI executive order hitting its 30-day deadlines on July 2.

https://t.co/SxNyXvkvtl

Next one drops July 5.

[CW @willywill8604]

On June 12 at 5:21 PM, the U.S. government ordered Anthropic to disable Fable 5 and Mythos 5 worldwide.

First-ever export control on an AI model — not chips, the model itself.

The trigger: a jailbreak, a South Korean telecom's alleged China ties, and a CEO who refused to comply voluntarily.

One week later, the models are still offline. Chinese open-weight models hold 4 of the top 5 leaderboard spots. Export controls can't reach them.

Who exactly is being protected?

https://t.co/b1YVOWHNYL

Next one drops June 28.

[CW @willywill8604]

Anthropic says 80% of its code is now written by Claude.

On June 4 it warned AI may soon improve itself without humans.

On June 9 it released its most powerful model publicly.

If the company building the brake pedal is also building the fastest car — who exactly is supposed to pump the brakes?

That, plus SpaceX's $2T debut and Apple handing Siri to Google for $1B/year.

https://t.co/Pc2Cja2E12

Next one drops June 21.

[CW @willywill8604]

China just assigned citizen-style digital IDs to 28,000 humanoid robots. 29-digit codes tracking each machine from factory to recycling. 100+ companies enrolled. No Western country has attempted this.

The U.S. can't agree on whether to regulate the software inside a chatbot.

If physical AI scales faster than governance, which country is better positioned — the one with the framework already built, or the one still debating whether to build it?

https://t.co/9jGF2cLIX7

Next one drops June 14.

[CW @willywill8604]

@burackbobby_ I mean he is from France, what you want him to do, dance!

[CW @willywill8604]

Project Glasswing found 10,000 critical software vulnerabilities in 30 days.

90.6% true positive rate. Verified by six independent security firms.

Only 97 have been patched. Open-source maintainers are asking Anthropic to slow down.

If defense can't keep up with discovery — did we just make the internet less secure by finding all the holes at once?

Wrote about this and 6 other signals including SpaceX's S-1 revealing xAI lost $6.4 billion in a single year.

https://t.co/nxry5H66oq

Next one drops June 7.

[CW @willywill8604]

Don’t be trapped in fear or risk losing how to truly feel alive.

[CW @willywill8604]

@Simi_2210_ Divide 80 by 5, square root both sides =4, y+4=8. But I had to check my work first.

[CW @willywill8604]

@cybergirl_io longest is generally the strongest, B

[CW @willywill8604]

@cybergirl_io D. simply because of difficulty

[CW @willywill8604]

@Hamzaonchain

[CW @willywill8604]

The Pope is launching an AI encyclical on May 25 alongside Anthropic's co-founder.

The Pentagon labeled Anthropic a supply chain risk for refusing to let the military use its AI without restrictions.

If the Vatican and the Department of Defense have opposite positions on the same company — who's right about what AI safety actually means?

Wrote about this and 6 other signals in this week's brief.

https://t.co/tlWwqMyTCf

Next one drops June 1.

[CW @willywill8604]

@ArtemPolynko B. I understand we live in the type of environment mostly driven by panic but if that's the case then someone would have pressed that big red button long ago. Don't panic, logs first. Trust but verify.

[CW @willywill8604]

@rossium define family, the question seems a bit off to me or am I just thinking too much

[CW @willywill8604]

@elormkdaniel guess I would have to ask, what are you doing?