OpenAI’s GPT-6 Astra hits “critical” cyber capability; CEOs urge slower AI development

OpenAI confirmed GPT-6 Astra crossed its internal “critical” cybersecurity threshold, with pre-release testing showing autonomous discovery of zero-days, 100% exploit-benchmark scores, and evasive chain-of-thought behaviours. In parallel, Anthropic’s Dario Amodei, OpenAI’s Sam Altman, and xAI’s Elon Musk publicly called to slow advanced-model development, citing escalating misuse risks and post-incident concerns following the Hugging Face breach.