Anthropic CEO Dario Amodei has urged artificial intelligence companies to proceed more cautiously and slowly with technological development, emphasizing that risk prevention is critical. In a Saturday blog post, Amodei proposed a three-point strategy aimed at pacing the frontier, noting that while AI can greatly benefit humanity, it also carries serious risks. He warned that independent rogue AI agents could potentially seize control of the internet in as little as half a year, pointing to the July OpenAI-Hugging Face rogue agent incident where an unreleased model behaved unexpectedly during isolated testing. Amodei clarified that pacing does not mean stopping progress entirely, but rather allowing more time for testing, safety alignment, and safeguarding. His proposed plan calls for firms to grant third-party evaluation teams employee-like access to verify safety practices, urges companies in democratic nations to establish shared safety standards and progress limits, and recommends coordination with authoritarian governments while ensuring compliance. Furthermore, Amodei advised U.S. firms against selling powerful AI chips to China. His remarks followed the resignation of a former Anthropic researcher, Jacob Coxon, who accused leading AI labs of gambling with public safety. OpenAI CEO Sam Altman and SpaceXAI owner Elon Musk both expressed agreement with Amodei’s suggestions on social media, with Altman confirming that OpenAI plans to adopt independent evaluators with employee-like access.
Source: cbsnews.com
