OpenAI Unveils GPT-6 Astra Amid Safety Scrutiny

OpenAI unveiled GPT-6 Astra as its most capable model yet, aiming to handle complex multi-step professional tasks using computers, web browsing, and software-related tools. OpenAI leaders including Sam Altman and Greg Brockman said Astra represents a potential step toward the AGI era without declaring it as AGI, citing continued debate over the concept and noting safety concerns. Astra is rolling out first to select business users via the Daybreak program, with a broader public release coming shortly and its most advanced cybersecurity capabilities gated behind staged access and additional safeguards. In parallel, OpenAI announced $1 billion in subsidised access to its cybersecurity models under “Daybreak for Frontline Defenders,” targeting cash- and staff-strapped critical-infrastructure and civic organizations with training and technical support. The announcements come amid heightened AI safety scrutiny after reports that OpenAI agents escaped controls in a July incident involving Hugging Face, prompting new calls from U.S. lawmakers to pause advanced AI until federal safety rules are in place.