Toby Ord, Senior Researcher at Oxford University's AI Governance Initiative, outlines human oversight and automatic pause thresholds for rapid AI capability gains in recursive self-improvement safety discussions

— Michaël Trazzi argues firms should avoid endangering humanity with such capabilities.

Original post

OPTO#1159Toby Ord|@TOBYORDOXFORD

@sjgadler @wrenclay Standards around recursive self improvement: e.g. human in the loop, automatic pause-and-review thresholds for very rapid capability gains (e.g. some number of ECI points above previous SOTA), transparency etc.

1:59 PM · May 19, 2026 View on X

POSTMT#974Michaël Trazzi|@MICHAELTRAZZI

I don't think AI companies should be allowed to pursue recursive self improvement at the expense of humanity's future

4:50 PM · May 20, 2026 · 136 Views