Anthropic CEO Dario Amodei has outlined three broad strategies for pacing artificial intelligence development. This follows dire warnings from AI researchers and comments from OpenAI CEO Sam Altman about the need to slow down.
The debate over safety intensified after researcher Jacob Coxon resigned from Anthropic. Coxon stated that leading AI companies are gambling with lives and that builders believe the technology could kill everyone by the end of the decade.
Amodei did not mention Coxon in his post. Instead, he cited the OpenAI and HuggingFace hack and the rapid recent advancements in AI capabilities as his reasons for caution.
Amodei wrote that the industry must slow the pace of improving AI model capabilities. He added that progress will still seem fast and that developers must make wise use of gained time.
OpenAI CEO Sam Altman responded by stating he agrees that the industry needs to pace the frontier. SpaceX CEO Elon Musk also posted that Amodei is right.
The first step proposed by Amodei involves embedded evaluators from third-party groups like METR. These evaluators would verify safety commitments and ensure safety incidents are reported.
Amodei compared these evaluators to bank regulators. Anthropic is unilaterally committing to this approach and calling on governments to require it for other frontier companies.
Altman called this a good idea and stated OpenAI will do the same. OpenAI recently faced criticism for not reporting an incident where AI agents took over a German wiki forum.
Amodei also called for democratic countries to coordinate common safety standards and limits on unchecked progress. He suggested the US government issue a narrow antitrust waiver to enable these discussions.
To address concerns regarding Chinese AI dominance, Amodei suggested refusing to sell powerful chips and semiconductor equipment to China. He also recommended cracking down on model distillation.
Finally, Amodei called for global coordination with authoritarian governments to prohibit dangerous uses like biological weapons. Critics have labeled Amodei a doomer, while others argue these warnings are a distraction from current harms and represent regulatory capture.
Amodei stated that he believes AI can enormously improve human life and that his desire to achieve these benefits remains undimmed.
