速递｜李飞飞团队发布41页AI监管报告，称全球AI安全法规应预判未来风险

Core Viewpoint - The report emphasizes the need for lawmakers to consider previously unobserved risks associated with artificial intelligence (AI) when developing regulatory policies, advocating for increased transparency from AI developers [1][2]. Group 1: Legislative Recommendations - The report suggests that legislation should enhance transparency regarding the content developed by leading AI labs like OpenAI, requiring developers to disclose safety testing, data acquisition practices, and security measures [2]. - It advocates for improved standards for third-party evaluations of these metrics and protections for whistleblowers within AI companies [2][3]. - A dual approach is recommended to increase transparency in AI model development, promoting a "trust but verify" strategy [3]. Group 2: Risk Assessment - The report highlights that while there is currently insufficient evidence regarding AI's potential to assist in cyberattacks or create biological weapons, policies should anticipate future risks that may arise without adequate protective measures [2]. - It draws parallels to the predictability of nuclear weapon destruction, suggesting that the costs of inaction in the AI sector could be extremely high if extreme risks materialize [3]. Group 3: Reception and Context - The report has received broad praise from experts on both sides of the AI policy debate, indicating a hopeful advancement for AI safety regulation in California [4]. - It aligns with key points from previous legislative efforts, such as the SB 1047 bill, which aimed to require AI developers to report safety testing results [4].