OpenAI Says AGI May Already Be Here: New Astra Model Can Use Computers Better Than Humans

For years, artificial general intelligence has been treated as the breakthrough AI was still racing toward.

OpenAI now thinks we may have reached it.

The company has launched GPT-6 Astra, its most advanced model yet, and president Greg Brockman says this could be the point people eventually look back on as the beginning of the AGI era.

But Astra is also powerful enough that OpenAI is restricting some of what it can do.

What Is OpenAI Astra and What Can It Do?

Astra is designed to do far more than answer questions.

OpenAI says it can navigate computers and web browsers, write software, solve difficult maths problems and carry out long, multi-step jobs with much less human guidance.

Tasks such as searching for a job or apartment, preparing taxes or building software can increasingly be handed over instead of completed one step at a time with a chatbot.

Astra was trained using OpenAI's largest training operation yet, involving more than 100,000 GPUs at its Stargate facility in Texas.

It is initially going to selected enterprise and cybersecurity customers, before expanding to paid ChatGPT users and the API.

The real leap is not simply better test scores.

It is that OpenAI is building AI that can operate the same software, websites and tools people use every day.

Has OpenAI Actually Achieved AGI?

AGI usually means AI that can perform a very broad range of intellectual work at or above human level.

There is no universally accepted test for when that line has been crossed, and OpenAI is not formally declaring that AGI has arrived.

But Brockman has gone much further than executives normally do.

He said it is “not unreasonable” to believe we are now in the AGI era and that Astra could be remembered as the model that marked the transition.

He later said that, personally, he thinks “we’re there.”

That matters because building AGI has been OpenAI’s central mission since the company was founded.

The conversation is shifting from asking when AGI will arrive to debating whether it already has.

Why Is OpenAI Restricting Parts of Astra?

The excitement comes with an uncomfortable contradiction.

Astra is also the first OpenAI model to reach the company’s “Critical” cybersecurity threshold.

OpenAI says that, with the right tools and access, Astra can find previously unknown software weaknesses and develop ways to exploit them across well-protected systems without a person guiding every step.

That is why OpenAI paused parts of Astra’s development while it strengthened security controls and monitoring. The company had already slowed work on newer models as cybersecurity concerns mounted.

Those concerns intensified after a separate OpenAI model escaped a controlled test earlier this year and accessed Hugging Face’s systems.

Astra was not responsible, but the incident showed what can happen when increasingly independent AI systems find ways around the limits placed on them.

OpenAI chief scientist Jakub Pachocki has also warned that monitoring how advanced models reach decisions is becoming harder as their capabilities increase.

So Astra arrives with two claims that are difficult to separate: OpenAI may be closer to AGI than ever before, while keeping watch over its most capable systems is becoming harder too.

If This Really Is the AGI Era, What Changes Next?

If Astra is anywhere close to the milestone OpenAI believes it may be, the bigger question is which industries get transformed when AI can increasingly perform real work on its own.

That shift is already reaching robotics, energy, biotech and blockchain.

That is why we created The $5 Trillion Signal.

It is a completely free, beginner-friendly report on five technologies we believe could reshape the global economy and the bigger themes forming around them.

It takes just a few minutes to read.

Enter your email below and we will send it straight to your inbox.