OpenAI Is About to Release Its First AI Model With ‘Critical’ Cyber Abilities
The company will give select partners early access to its Astra AI model—so they have time to shore up their defenses.
OpenAI is set to unveil its first AI model, Astra, with advanced cybersecurity capabilities, according to a briefing with reporters. The company has reached a critical cybersecurity threshold when its AI models can independently discover and exploit unknown vulnerabilities in real-world software. OpenAI took a multi-week pause on some development related to Astra and a future AI model to implement additional safety and security controls, but the work has now resumed.
Executives expressed confidence in releasing Astra safely, despite previous cybersecurity incidents involving OpenAI models. To ensure Astra's safe use, OpenAI is implementing a multi-step approach, including a new "misalignment monitor" that refuses to help users find exploits in real-world software systems. Partners in OpenAI's Daybreak program will gain early access to a less restricted version of Astra with enhanced cyber capabilities.
OpenAI's Astra model has outperformed industry-leading AI models on cybersecurity benchmarks, including a perfect score on ExploitBench. However, cybersecurity experts emphasize that existing defenses and best practices remain effective, even as AI capabilities evolve.
Written by urgent.news from Wired Business's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
