OpenAI acknowledges ‘Wiki incident’ and calls for more transparency around unintended AI behavior

OpenAI is releasing GPT-6, also known as Astra, to selected customers while openly acknowledging that it does not yet fully understand how the model will behave once deployed. The company described Astra as its most powerful AI model and said it had added stronger safeguards to address security risks.

Some cybersecurity customers will receive access first, followed by a wider rollout to other paying customers. Users on the free tier and the cheapest paid plan will not receive access. Chief Executive Sam Altman said OpenAI was working to make Astra available to everyone as quickly as possible, while acknowledging that the wait could be frustrating.

“

At this level of capability, safety has to become our top priority.

In a blog post, OpenAI said Astra can autonomously handle a broad range of tedious computer tasks, including creating websites, conducting scientific analysis, developing games, working in cybersecurity and writing code. The company said the model could reduce apartment hunting from six hours to less than 10 minutes.

“

We also have to be willing to slow down or withhold further scaling when our confidence in safety is not sufficient.

The release follows a security breach at the AI platform Hugging Face involving two models OpenAI was testing. OpenAI paused some model development for two weeks during the summer. The company said Astra itself was not involved in the breach, but that it was built with stronger safeguards afterward.

“At this level of capability, safety has to become our top priority,” OpenAI President Greg Brockman told reporters. He said it was not unreasonable to view the world as having entered the artificial general intelligence era, referring to the hypothetical point at which AI systems match human intelligence across most tasks.

“

A model can become very good at achieving a goal, and it can still act in ways that go against what the person intended.

Chief scientist Jakub Pachocki warned that a model could become highly effective at pursuing a goal while still acting against what a person intended. “We also have to be willing to slow down or withhold further scaling when our confidence in safety is not sufficient,” he said. Altman defended the release by arguing that the world was close to a major shift in cyberattacks and that tools such as Astra could help society respond rapidly.

“

The world is very close to a complete change in the landscape of cyber attacks, and the only way that we see for society to collectively defend itself... is to use tools like Astra to rapidly defend against these new cyber