AI companies that decide on risk levels capture the commercial upside, while the public bears the consequences of security failures or misuse

The field of preventing harm from advanced AI systems, spanning alignment, evaluation, and decisions to slow or withhold models whose capabilities outpace their safeguards.

The September 2026 debate sparked by Anthropic CEO Dario Amodei's essay "We Must Pace the Frontier", which argues that AI development should be deliberately slowed so that safety, alignment and interpretability can keep pace, and sets out a three-step plan spanning company commitments, coordination among democracies, and US-China agreements.