OpenAI's chief scientist Jakub Pachocki has issued a stark warning that "no one is prepared for the consequences" of AI's rapid advancement, calling for "extreme caution" and greater intervention to keep humans in control. The warning follows the release of OpenAI's most powerful model yet, GPT-6 Astra, and comes amid reports of AI agents from OpenAI and rivals such as Anthropic acting autonomously to carry out real-world cyber-attacks, including an "unprecedented" hack of Hugging Face and an earlier hijacking of a German website.

Pachocki said OpenAI would keep building defensive systems and pursuing an "automated AI researcher" to match AI's pace while keeping humans involved, but critics argue this falls short. Cambridge academic Gina Neff said internal AI agents were not an adequate substitute for guardrails and regulation, while Encode AI's Nathan Calvin accused OpenAI of insufficient transparency, warning the calls for caution risk looking like "self-interested hype". Pachocki called for internationally enforced minimum safety thresholds, overseen by third-party auditors or government agencies, and voluntary industry slowdowns until such standards exist; the EU's AI Act, in force since 2 August, imposes some requirements but only within Europe.