Researcher argues that building safe superintelligent AI is only feasible if ASI development is globally prohibited to prevent competitive pressures.

The article critiques various proposed approaches to safe ASI, including aligned superintelligence, tool AI, and non-agentic systems, calling most suggestions unworkable.

The author argues that even with a working safety plan, competitive dynamics could force unsafe ASI development before safe versions are completed.

The core argument suggests global ASI bans are necessary to eliminate the race-condition problem that makes voluntary safe development impractical.