Anthropic's Leaked Docs Reveal Plan to Beat China with AI

Internal documents from Anthropic, a leading AI company, were made public due to a book copyright lawsuit. These papers, written in 2021 and 2023, reveal the company's early plans to build a super-smart AI assistant that could write and fix software just like a human engineer. The goal was to create a "proto-AGI"—a stepping stone to artificial general intelligence that could handle many tasks.

One surprising detail: an employee suggested training the AI on Chinese text specifically to slow down Chinese AI companies. The idea was that by capturing markets in countries along China's Belt and Road initiative, Anthropic could tip the geopolitical balance. This shows how AI development is not just about technology but also about global power.

The documents also discuss safety. One employee worried that if Anthropic focused on making AI safe in one area (like language) but powerful in another (like coding), the safety work might not keep up. They proposed a principle: capabilities and safety research should grow together. This is important because as AI gets more powerful, we need to ensure it doesn't cause harm.

Finally, a 2023 slideshow outlined product visions for 2025, suggesting Anthropic aim for transformative AI but break it into achievable steps, like SpaceX did with Mars. This approach helps companies stay grounded and profitable while chasing big dreams. Overall, these leaks give a rare look at how AI companies balance ambition, safety, and competition.