Report Raises Security Concerns Over Moonshot AI's Kimi
An internal memo from iapsAI has raised concerns about Moonshot AI's "always-on" browser agent, Kimi Claw. The memo suggests the agent, which can see and act on all user activity, routes data through infrastructure subject to China's National Intelligence Law. The author describes the potential for data scraping as a far greater risk than that posed by TikTok.
Beijing-based Moonshot AI has seen a meteoric rise, achieving a valuation of $4.3 billion after a $500 million Series C in January 2026, with significant backing from tech giants like Alibaba Group and Tencent. The company's latest funding round in February 2026 is targeting a valuation as high as $12 billion, fueled by exploding investor interest and surging overseas revenue for its Kimi models. Kimi Claw is an "agentic browser," a class of AI that moves beyond passive assistance to take autonomous action on a user's behalf—navigating websites, filling forms, and executing multi-step workflows without direct intervention. Architecturally, these agents require deep integration into the browser's operational framework, interpreting high-level goals and deconstructing them into a sequence of executable web tasks. The security concerns are rooted in China's comprehensive data governance framework, which includes the Data Security Law (DSL) and the Personal Information Protection Law (PIPL). These laws classify business data based on national security relevance and place strict controls on cross-border data transfers, requiring government approval for sharing data with foreign judicial or law enforcement bodies. For consumer AI products, the core challenge is a direct trade-off between agent capability and user trust. An "always-on" agent with full visibility into user activity offers maximum utility but requires extensive permissions that create significant privacy risks. Designing interfaces that build trust by clarifying what an AI is doing, why, and how reliable it is, becomes a critical differentiator in a market where users are wary of data leakage. Orchestrating such powerful agents at scale often involves a supervisor-agent architecture, where a primary agent decomposes complex tasks and delegates them to specialized sub-agents. Frameworks like LangGraph and open standards such as the Model Context Protocol (MCP) are emerging to manage these intricate, stateful interactions and ensure more predictable, reliable agent behavior. This isn't the only recent scrutiny on the company. In February 2026, US-based AI developer Anthropic accused Moonshot AI, along with other Chinese labs, of using fraudulent accounts to make millions of queries to its Claude model. The alleged aim was to "distill" Claude's capabilities to train their own models, raising questions about the provenance of the underlying tech. Moonshot AI's strategy with Kimi Claw reflects a broader market shift from foundational models to integrated agent platforms. By bundling a powerful model with a massive library of pre-built skills and cloud infrastructure, they aim to make developer experience and time-to-value the key competitive battleground, challenging the fragmented, DIY approach of wiring separate tools together. The company has also been active in the open-source community, releasing the weights for Kimi K2, a 1-trillion-parameter Mixture-of-Experts (MoE) model. More recently, it launched Kimi K2.5 and Kimi K2 Thinking, updates designed specifically for advanced reasoning and agentic tasks, including the ability to autonomously execute hundreds of sequential tool calls.
