Investing.com — OpenAI is halting some internal development of its next-generation artificial intelligence model after preliminary tests suggested the software could autonomously execute sophisticated cyberattacks.
The San Francisco-based startup said Thursday that its upcoming model, code-named Astra, demonstrated advanced coding and hacking skills that may cross its internal “Critical” risk threshold. Under OpenAI’s self-imposed safety framework, an AI system reaches this classification if it can independently discover “zero-day” software vulnerabilities or launch end-to-end attacks on secure networks without human direction.
In response to the capability jump, OpenAI is restricting Astra’s development to heavily guarded, isolated environments.
At a Glance: OpenAI’s Security Overhaul for ‘Astra’
The preemptive lockdown underscores the growing tension in Silicon Valley between the race to commercialize increasingly powerful AI and the imperative to contain it. Previous iterations of OpenAI’s technology, including GPT-5.6-Sol, maxed out at a “High” risk rating.
OpenAI clarified that Astra remains unreleased and was not involved in recent high-profile AI security breaches, such as the recent Hugging Face exploit. The company framed the pause as evidence that its internal safety guardrails are working as intended, catching dangerous capabilities before the technology is deployed to the public or enterprise clients.
Related articles
OpenAI flags critical cyber capability risk in upcoming ‘Astra’ model
Celsius jumps 12% as Rockstar founder builds 4.7% stake, demands CEO ouster
Atlassian results seen as ‘game changer’ for AI kills software worries