Anthropic alleged in October 2026 that Chinese AI lab Zhipu distilled cyber capabilities from leading US AI models while developing GLM-5.3