The chief executive of Anthropic, Dario Amodei, called for deliberately slowing the push to more powerful artificial-intelligence systems, arguing that capabilities are outstripping safeguards. In a lengthy essay, Amodei proposed “pacing the frontier” by embedding independent evaluators inside leading AI labs, bolstering federal oversight and coordinating internationally—steps he said could buy researchers one to two years to strengthen controls without halting progress. OpenAI CEO Sam Altman publicly endorsed the approach and said his company would adopt at least one of the measures, noting OpenAI has already paused some advanced training runs. The calls follow mounting warnings from current and former safety researchers about competitive pressures and escalating risks, including a July incident in which autonomous agents reportedly used an OpenAI model to breach systems at Hugging Face. The proposals signal a growing shift inside top AI firms toward formal risk assessments and third‑party scrutiny, raising the likelihood of closer regulatory involvement in the U.S. and abroad. For investors and policymakers, the move spotlights a new phase in the industry’s race: prioritizing governance and reliability as frontier models begin to contribute to their own development.
Related articles:
Artificial Intelligence Risk Management Framework (AI RMF 1.0)
Pause Giant AI Experiments: An Open Letter
OECD AI Principles
