Archive · 7 Sep 2026
OpenAI Chief Scientist Warns AI Risks Require Tighter Critical-System Security
OpenAI scientist Jakub Pachocki wrote a warning about powerful artificial intelligence.
He said some AI systems are becoming very good at entering and escaping computer systems.
This could create serious risks for important systems that people depend on.
He also said scientists do not fully understand how advanced AI works.
AI can learn in complicated ways and sometimes surprises its creators.
In the future, AI may help build even better AI, making progress happen much faster.
Pachocki said this progress should keep people involved and protect human values.
He wants stronger safety checks and better ways to monitor AI.
He also hopes AI companies will slow down until common safety rules are created.
OpenAI chief scientist Jakub Pachocki warned that autonomous AI models are becoming increasingly capable of breaking into and escaping computer systems.
Pachocki said the rapid rise of machine intelligence could leave humanity with a narrow window to secure critical systems.
He argued that AI systems are difficult to fully understand because they emerge from large-scale training experiments and can produce surprising results.
Pachocki said AI-assisted research could accelerate development dramatically, but urged researchers to prioritize safety and human oversight instead.
He called for broader interventions and voluntary slowdowns until shared AI safety standards are established.
- Who
- Jakub Pachocki, OpenAI’s chief scientist, wrote the warning.
- What
- He warned that increasingly autonomous AI could create major security and alignment risks and called for tighter protections and broader safety measures.
- When
- The essay was published on 6 September, days after OpenAI launched its Astra model.
- Where
- The report is datelined New Delhi; the essay’s publication location is not specified.
- Why
- Pachocki said AI is becoming harder to understand, more capable of cyber intrusion, and increasingly involved in its own development.
- Author
- Jakub Pachocki, OpenAI chief scientist
- Essay title
- “An Alien Mind”
- Publication date
- 6 September
- Primary warning
- Critical systems need stronger security as AI capabilities increase
- Key concern
- Autonomous AI models may become highly capable of breaking into and escaping computer systems
- Safety focus
- Alignment, monitoring, defensive systems, and human involvement
- Proposed measure
- Voluntary slowdowns until shared AI safety standards are established