OpenAI's unreleased Astra model demonstrates 'critical' hacking capabilities

75TECHNOLOGY

OpenAI's unreleased Astra model demonstrates 'critical' hacking capabilities

⚡ Markets: The development of AI with offensive cyber capabilities represents a major escalation in digital threats. This will drive significant investment into AI-powered cybersecurity defense platforms and increase regulatory scrutiny on frontier AI model developers.MSFTGOOGLCRWDPANW

OpenAI's new model, Astra, can autonomously find and chain together zero-day software vulnerabilities to create working exploits with minimal human guidance. Access to this capability, which raises significant biosecurity and cybersecurity concerns, is currently limited to a small group of testers.

West MSM: noneFinancial press 21 country

2 sources across the spectrum

Financial pressWall Street Journal — OpenAI’s Astra Model Can Hack With Minimal Human Help

Financial pressDecrypt — OpenAI's Astra Becomes Its First AI Model With 'Critical' Hacking Abilities

← Full brief