Dario Amodei, chief executive officer of Anthropic, has urged companies working in artificial intelligence (AI) to slow the pace at which they develop model capabilities. Amodei argued that a slower tempo would give firms the necessary time to properly tune and secure their models and to allow independent experts to review those steps.
In an essay shared on X, Amodei wrote that “the progress will still feel fast, and we should use the gained time wisely.” His appeal followed the publication on Thursday of a threat-analysis report by Anthropic that detailed how multiple market players had used the company’s Claude models for abusive purposes.
According to the report, Claude models were exploited in activities ranging from weapon development and cyber operations to surveillance and fraud. Anthropic alleged that Moonshot, a leading Chinese AI company, secretly forwarded requests it received to Anthropic’s Claude chatbot and then used Claude’s responses to develop its own language model. The report also named Deepseek and Xiaomi as engaging in similar behavior.
Concerns about harms from AI escalated this week after Jacob Coxon, a researcher at Anthropic, resigned, stating that “those who are building AI seriously believe it could kill us all by the end of the decade.”
Amodei clarified that he is not calling for a halt to model training or technical progress, but for companies to devote sufficient time to make models safe and to have independent experts verify those safety steps. As part of the three-step framework he proposed, Anthropic would place permanent, independent monitors at frontier AI companies; these monitors would be given access to relevant tools and to internal risk-assessment processes.
The proposed measures aim to ensure that firms do not only accelerate capability growth but also introduce new capabilities in ways that are safe and transparent. Amodei argued that a moderated pace combined with independent oversight could provide the “gained time” needed to reduce potential misuse and harms.
Why this matters
The call and the threat-analysis report reflect growing worries within the industry that rapid expansion of model capabilities carries significant safety and ethical risks. Anthropic’s proposals are likely to fuel further debate over how frontier AI research and development should be regulated and overseen.
(Information based on Anthropic’s essay and threat-analysis report; reporting by Reuters.)
