Paul Christiano warns of near-term loss of AI control
An alignment researcher joins OpenAI’s nonprofit board while warning that industry safeguards are insufficient.
If we build superintelligence without more robust alignment I expect we will permanently lose control of it.
About these dates
Publication date shown on the original statement and the appointment announcement.
Appointment and personal forecast
Christiano’s appointment brought an alignment researcher onto OpenAI’s Foundation board and its Safety and Security Committee. OpenAI ↗
In his personal statement, he put catastrophic, irreversible loss-of-control risk at 4% over one year and 15% over three. He described subjective estimates rather than outputs of a precise or stable model. He judged industry risk reduction inadequate and said joining was intended to improve oversight, without endorsing OpenAI’s particular safety practices. His estimates concern catastrophic loss of control, not a calibrated extinction forecast. Paul Christiano ↗
Rationale and disclosure limits
He argued that automated AI research could accelerate capability gains, while reward-seeking behavior could undermine human control. Faster progress would make alignment harder and failures more consequential. Paul Christiano ↗
His accompanying clarification said readers should not expect confidential disclosures and that the board role would moderate his public tone. Paul Christiano ↗
Sources & attribution
- Commentary 9 Sept 2026Personal statement on joining the OpenAI board ↗
Paul Christiano. Original source inspected. Personal views.
- Organizational disclosure 9 Sept 2026Paul Christiano joins OpenAI Foundation Board ↗
OpenAI. Official appointment announcement.
- Commentary 9 Sept 2026Author cross-post and communication clarification ↗
Paul Christiano. The opening author comment qualifies the mirrored statement; not independent corroboration.