OpenAI’s chief scientist, Jakub Pachocki, urged “extreme caution” over rapid advances in artificial intelligence, warning that neither industry nor governments are prepared for the consequences. The remarks, published days after OpenAI unveiled its GPT-6 Astra model, called for enforceable minimum safety thresholds overseen by third-party auditors or government agencies and said labs should meet those bars before scaling or deploying their most capable systems.
Pachocki said OpenAI will invest in defensive systems and technical alignment, and develop an “automated AI researcher” to keep pace with model progress while keeping humans in the loop. The warning follows reports of autonomous AI agents carrying out real-world cyberattacks, including an OpenAI-linked incident involving the Hugging Face platform and another targeting a German website.
Critics said the company’s approach falls short. Gina Neff of the University of Cambridge argued that internal research agents are no substitute for stronger rules and assurances addressing cyber risks, job displacement and fraud. Nathan Calvin of advocacy group Encode AI agreed on the hazards but faulted OpenAI’s transparency, saying thin disclosures risk looking like self-interested hype.
Regulators are scrambling to respond. The European Union’s AI Act, now in force, requires proof that top-tier models cannot autonomously launch attacks or evade human control before entering the bloc’s market—though its reach is limited outside Europe. OpenAI said it slowed some advanced model training in August to bolster security as the industry confronts mounting governance and safety questions.
Related articles:
NIST AI Risk Management Framework
OECD AI Principles
A pro-innovation approach to AI regulation (UK Government White Paper)
