Do Son Kimi Work's desktop feedback tool silently uploads raw records from five recent agent sessions alongside diagnostic logs, with no disclosure or opt-out option.
Moonshot AI
Moonshot AI open-weights Kimi K3, the first 3T-class open model, and reveals training agents that triggered kernel panics, prompting a microVM sandbox rebuild.
The Distillation War: Anthropic Accuses DeepSeek and Chinese Rivals of ‘Industrial-Scale’ Data Theft
The Distillation War: Anthropic Accuses DeepSeek and Chinese Rivals of ‘Industrial-Scale’ Data Theft
Anthropic claims DeepSeek and other Chinese AI firms used 24,000 accounts to "pilfer" Claude's data. Inside the high-stakes battle for LLM training secrets.
Moonshot AI open-sourced its Kimi K2 model, a 1T-parameter MoE AI, showing exceptional performance (outperforming Grok) and prompting Perplexity AI to use it for post-training.