Synthetic Minds | Amazon Phoned the Treasury and Anthropic Went Dark

Synthetic Minds | Amazon Phoned the Treasury and Anthropic Went Dark

The Synthetic Minds newsletter offers short daily insights to get you thinking. If you enjoy it, please forward. All signals are powered by Futurwise. If you need more insights, subscribe to Futurwise and get 25% off for the first three months!

I have just launched the Intelligence Age Scorecard! It will help you understand how ready your organization is for the Intelligence Age.

Today’s topic: AI & Automation

One Phone Call Switched Off Anthropic

The kill switch on the world's most powerful AI was pulled by one phone call. Amazon told the US Treasury that the AI it funds and hosts had a flaw. The AI went offline for everyone until Anthropic finds a solution to comply with the decree.

Amazon plays four roles at the same lab: investor, landlord, whistleblower, and competitor. Four jobs in one chair.

Amazon researchers tricked Anthropic's most powerful AI into giving up information that could help a cyberattack. Amazon CEO Andy Jassy phoned the Treasury Secretary with the finding.

The Commerce Department then ordered Anthropic to switch off both of its best AI models. The order covers any foreign national anywhere in the world, including Anthropic's own non-American employees.

The government gave Anthropic a choice. Fix the flaw or pull the model. Dario Amodei refused and pulled the model.

Anthropic says the flaw is small, that other AI models have the same one, and that OpenAI's GPT-5.5 can be tricked the same way.

That's the news. Here is the signal.

Amazon sells its own AI. The Nova family of models runs on the same Amazon cloud Anthropic runs on, and Amazon prices Nova below Claude.

When Claude's best models are offline, the customer staying on Amazon's cloud has one less option to choose from. One of the remaining options is Amazon's own.

The argument that the frontier AI lab had been crossed into public infrastructure named the actors. Five regimes sitting around one company. One of those regimes has shown it can switch the company off without warning, on a tip from another regime that benefits when it does.

Europe has noticed. A scenario by European researchers called Europe2031 lays out where this ends:

Most of Europe gets pushed into a second tier. Limited access to the best AI, capped at a quarter of the supply. Revocable at the next phone call from a Cabinet secretary.

The open-source, Chinese, AI you can download is four months behind the best American AI. Four months. Squeeze the American AI too hard and the world will switch, and Chinese labs already hold four of the top five free-AI positions globally.

The last time the world locked down dual-use technology, it took 33 countries and nearly three years of negotiation. This took one phone call from one CEO to one Cabinet secretary.

The question your board should debate is no longer which AI you bought. It is who can switch it off, and which of your employees will be cut off when they do.

The Intelligence Age Scorecard

The world's most powerful AI was switched off for everyone outside America in one afternoon, on a tip phoned in by the company that funds it, hosts it on its cloud, and sells AI that competes with it. Are you still watching which AI you bought, or already verifying who can switch it off and which of your employees will be cut off when they do?

Take the Intelligence Age Scorecard to benchmark your readiness for the next two quarters, and the next five years.

If this newsletter was forwarded to you, you can sign up here.

Thank you.
Mark