Anthropic wants caution with AI development

Analysis of AI development is ongoing.

The topic of AI is currently being discussed.

Monday, June 8, 2026, at 4:40 am

Anthropic wants to slow down AI development.

Anthropic Inc. has called for a global freeze on AI development so humanity can catch up.

However, this did not actually happen.

In a blog post, Anthropic explained that Claude is writing its own code and proposing experiments.

The blog suggests slowing down AI development if possible, with global coordination.

This might be a good thing if it were possible, with a global mechanism including China and Russia.

Building defenses against AI is important.

Anthropic has built a new AI model called Mythos, which is very good at computer security tasks.

Mythos was not released to the public, but was given to a select group to prepare defenses.

This group is called Project Glasswing, and it includes some Australian companies.

It is not known how these companies were chosen.

Anthropic's CEO said Project Glasswing is to help democracies stay ahead of autocracies.

The government is under pressure to respond to possible AI risks.

This is similar to a conference in 1944 where countries met to set up a new economic system.

However, Project Glasswing is a private sector project, not a government one.

Another relevant event was the invention of the first computer, Colossus, in 1944.

There is a direct line from Colossus to Mythos, and from one kind of D-Day to another.

After the war, the US defense department funded the development of digital computing.

Later, technology became a private sector race between American companies.

Now, AI is the culmination of Colossus, with Anthropic, OpenAI, and SpaceX leading the way.

China has caught up with US technology, and Dario Amodei is protecting democracies.

The AI bubble has taken off.

Project Glasswing and the withholding of Claude Mythos were announced on April 7.

The stock market AI bubble started again on March 30, and has been growing since then.

There was no specific event that started this turnaround, just a renewed recognition of AI's importance.

AI is going to cost more than expected, and the infrastructure build will have to be huge.

Semiconductor shares have gone up the most this year, with the big three up sixfold in six months.

Shares in companies building semiconductors have gone up due to AI demand.

Will the bubble burst this year?

It's both different and the same this time.

The concentration of the bubbling sector is the same as before past crashes.

But there's never been a boom like this before, with real earnings and capital spending supporting prices.

The question is whether AI will become a low-margin utility business like electricity.

Providing data center compute for AI is becoming a utility business.

The business of making and selling AI tokens is just getting started, and seems profitable for now.

The product is called inference, which is the operational phase of AI after training.

In inference, the trained model takes input and computes output in the form of tokens.

How will the AI business develop?

It's likely to develop into three layers: GPU vendors, cloud providers, and LLM APIs.

RAG frameworks, orchestration, and domain-specific models will provide higher margins.

Vertical products that embed LLMs will have attractive unit economics.

But base LLMs like Claude and ChatGPT could become low-margin utilities.

There will be higher margins in AI-enabled products, such as agents and robots.

China is moving into the AI industry.

China has regulated emotional attachment to humanoid AI robots.

This is the first attempt by any government to deal with AI emotional dangers.

The new law contains provisions to prevent emotional manipulation and dependency.

Good luck trying to police it!

The big hyper-scalers are doing massive IPOs this year, but may become low-margin utilities.

The number one issue for investors is when they will realize AI is just another product.

Alan Kohler is a finance presenter and columnist.

Monday, June 8, 2026, at 4:40 am

Promotion

Prática de escuta

Prática de escuta