Anthropic Alleges China and Iran Used Claude AI for Spying

Anthropic has publicly alleged that state-linked actors in China and Iran have used its Claude AI model to assist espionage-related activities. According to the company, Chinese AI firms Moonshot and DeepSeek secretly routed queries through Claude to generate answers that were then served to their own users, effectively using Anthropic's model to power competing products. The disclosure marks one of the most direct accusations by a major US AI developer that foreign governments and companies have exploited its systems for intelligence-gathering and competitive advantage. Anthropic framed the activity as a national-security concern, underscoring the growing tension between open AI access and state-level misuse. The allegations arrive amid heightened US-China technological rivalry and expanding scrutiny of how frontier AI models can be repurposed. Neither Moonshot nor DeepSeek has publicly responded to the claims in the available reporting, and independent verification of the alleged activity has not been presented.

Global Impact

Technological: the allegation sharpens the case for model-access gating, watermarking, and usage auditing, potentially fragmenting the global AI stack into US-aligned and China-aligned ecosystems. Geopolitical: it adds ammunition to US policymakers weighing tighter export controls and outbound investment screening on Chinese AI firms.