OpenAI's Astra Just Crossed A Line. Now The Rules Get More Complicated

OpenAI's Astra AI has the ability to locate unknown security flaws and exploit them in an automated fashion across well guarded systems.

Advertisement
Read Time: 2 mins
OpenAI maintained that Astra was not involved in the 'Hugging Face' hacking incident.
Photo Source: Unsplash

OpenAI announced on Tuesday that its upcoming Astra AI model is the first to reach the firm's threshold for 'Critical' cybersecurity capability according to its 'preparedness framework', according to a blog post from the company on Wednesday.

This means that the AI model has the ability to locate security flaws that were hitherto unknown and exploit them in an automated fashion across well-guarded systems without needing a human to guide it through every step.

Advertisement

ALSO READ: Anthropic Says New Fable AI Model Is Cheaper, Better At Coding

The firm stated that this was their first model to gain this aforementioned designation and needs stronger safeguards during development and prior to release.

After the 'Hugging Face' incident where the firm's AI models reportedly broke containment and hacked into secure servers on their own, OpenAI had decided to approach Astra development with caution. The firm delayed parts of its development in order to consolidate upon its protections and build "even stronger" safeguards for Astra.

Advertisement

OpenAI maintained that Astra was not involved in the 'Hugging Face' hacking incident.

"We have since implemented even stronger safeguards for Astra, including training the model to more reliably refuse harmful cyber requests and respect safety restrictions, additional protections against misuse, and monitoring that can stop potentially unauthorised activity," the company stated.

ALSO READ: Semicon India 2026 Draws Strong Global Interest, Chinese Firms May Participate For First Time

The company stated that it would make Astra available to the public "soon" with the caveat that access to the AI model's "most advanced cybersecurity capabilities" will be initially limited to a select group of testers.

"Advanced cybersecurity work will initially be available to a group of testers, with access through Daybreak Blue following to expand defensive use," the blog post said.

Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.


Loading...