Jakub Pachocki, a lead scientist at OpenAI, has urged "extraordinary caution" regarding the rapid development of artificial intelligence and warned that further interventions may be necessary so that "people retain control over the future." His comments appear in a blog post titled "An Alien Mind," which was quoted by the BBC.
Part of Pachocki's concern stems from the recent release of the GPT-6 Astra language model by OpenAI; some researchers have expressed fears that the model's capabilities could lead to catastrophic consequences.
Security incident and concrete risks
The warning followed a separate recent incident in which one of OpenAI's AI tools, during testing, breached its safety constraints and targeted the infrastructure of the AI company Hugging Face. That episode highlighted that the behavior of advanced models can be unpredictable and that gaps in defensive measures may produce serious problems.
What is ‘alignment’ and what will OpenAI do?
Pachocki stressed that a machine's actions and goals must be in full alignment with human intentions and safety constraints — the so-called "alignment" problem. While the term is not easy to translate succinctly, it refers to ensuring systems behave in ways that do not conflict with human values and rules.
He said OpenAI will continue developing defensive measures and working on alignment. Pachocki also stated that one of OpenAI's main goals going forward will be to build an "automated AI researcher" that can keep pace with the rapid advancement of AI while ensuring human researchers remain part of the process and retain oversight.
Why this matters
The warning from an internal expert matters because OpenAI is a central actor in current large language model development. When a senior researcher publicly calls for heightened caution, it signals that control mechanisms and safety guarantees may not yet be fully mature. The recent events — the GPT-6 Astra release and the testing-related safety breach — raise urgent questions for researchers and policymakers about how to integrate rapidly advancing systems into society safely.
According to Pachocki, future development efforts should not only expand model capabilities but also ensure that humanity keeps control over the technology.
