China's AI push, model safety concerns, and legal AI advances

China's tech giants are quietly building their own AI models, moving away from reliance on external providers. Companies like Meituan and Xiaohongshu are integrating custom AI systems into their services, a trend that accelerates specialized AI development within the country's consumer tech ecosystem. This domestic focus contrasts with the global spotlight on major AI labs, suggesting a strategic shift towards self-sufficiency in AI for everyday business applications.

Meanwhile, the industry grapples with the dual-use nature of advanced AI. Z.ai has temporarily withheld the weights for its new GLM-5.3 model due to its unexpectedly rapid development of offensive security capabilities, a move mirroring caution from OpenAI and Anthropic. This decision highlights growing industry-wide concerns about the release strategies for powerful open-weight models. On a more constructive note, Harvey unveiled Harvey Tenet, a model built on Kimi K3 and trained with Fireworks, which sets new state-of-the-art results on legal benchmarks, potentially accelerating AI adoption in the legal sector.