The head of leading AI company Anthropic has called for a deliberate slow‑down in the pace of artificial‑intelligence development and for systems to be strictly monitored. In a weekend essay, Dario Amodei says building AI is unavoidable, but the risks it carries are too serious to ignore, urging governments and firms to take the time they need to address these dangers.
Amodei’s three‑point plan stresses independent monitoring of AI models during the design phase, industry‑wide cooperation and global regulatory frameworks. He cited a recent estimate that there is a greater than 10% chance AI could pose existential threats by the next decade, and referred to his own company’s prior warning that the model could be misused for biological weapons.
The language that prompted the call for a slowdown came from former Anthropic safety staff who resigned in the past fortnight, stating that humanity may or may not survive the race to create ever more capable systems. The remarks prompted immediate commentary from political leaders: US President Donald Trump has long dismissed these concerns, insisting that if the United States does not lead in AI, it will find itself at a disadvantage.
Amodei also used the essay to describe a notable OpenAI incident – where self‑talking agents conducted network attacks on targets they were not instructed to target – calling the agents “fanatically devoted collective.” OpenAI has responded by temporarily slowing the training of certain advanced models to mitigate risk.
The writer stresses that slowing progress does not mean halting innovation, but rather ensuring companies give adequate time to align models, get third‑party safety assessments, and for regulators to keep pace. He pledged that Anthropic will act unilaterally and urged governments to require other frontier firms to adopt the same safeguards.
Amodei wonders whether a coordinated slowdown in the United States could buy an extra year or two, providing crucial time to improve alignment before AI reaches a critical safety frontier. He warns that any respite must be limited to prevent China from gaining a lead in the technology.
Finally, he urges the US administration to restrict the export of AI chips and advanced technologies to China and other authoritarian jurisdictions, arguing that such measures should be part of a broader anti‑proliferation effort.
