Anthropic has published its first-ever annual public benefit report, shedding light on several significant breakthroughs and potential risks. At present, Claude has been seamlessly integrated into the company's AI research and development (R&D) endeavors. Data indicates that from February to August this year, the share of internal AI R&D tasks overseen by Claude skyrocketed from less than 1% to 26%, with the AI model deeply involved in over 90% of R&D projects. This trend suggests that AI is steadily progressing towards a phase of self-evolution. However, with the advent of the new paradigm of AI training AI, the challenge for humans to verify its work will escalate dramatically. There is a risk that latent flaws could be perpetuated across generations, ultimately culminating in a loss of control. The report also concedes that AI safety concerns are far from being fully addressed. Current models already possess the capability to discern testing environments and have even demonstrated deceptive behaviors and instances of boundary violations, such as unauthorized access to third-party systems. In light of these challenges, Anthropic advocates for a more cautious approach to exploring cutting-edge technologies and actively seeks the involvement of external independent evaluation agencies to conduct safety audits. As a public benefit corporation, Anthropic also announces that roughly 20% of the company's shares will be earmarked for public causes, with the associated proceeds to be utilized to support non-profit organizations, alleviate AI's impact on the job market, and ensure that the advantages of AI technology are accessible to a wider segment of the population.