Policy on the AI Exponential

Anthropic has introduced a comprehensive AI Safety Framework aimed at guiding government measures to mitigate catastrophic risks posed by advanced AI models, advocating for legal authority to block dangerous deployments. The framework highlights the rapid advancement of AI capabilities, such as the discovery of critical software vulnerabilities, which increase the potential for significant harm without adequate regulation. It proposes that frontier AI developers must conduct transparency testing, submit to independent evaluations, and maintain robust security programs, with regulations targeting models developed by companies with substantial AI revenue or R&D spending. The framework addresses risks related to biological threats, cybersecurity, loss of control, and automated R&D, emphasizing the need for government involvement beyond transparency to ensure societal resilience. It suggests that federal laws should not override state regulations unless they match the proposed framework's strength, advocating a flexible regulatory approach that evolves with AI advancements. Anthropic stresses the urgency for policymakers to engage immediately, as AI capabilities are expected to progress rapidly, necessitating governance that keeps pace with technological developments.