TL;DR
China's top internet regulator has opened data security probes into DeepSeek and Moonshot AI, alleging both startups routed sensitive user data through a foreign model, Anthropic's Claude, in violation of Chinese law.
What happened
- China's Cyberspace Administration (CAC) launched formal data security investigations into DeepSeek and Moonshot AI on September 23, 2026.
- The alleged violation: both companies routed sensitive user data through Anthropic's Claude, a US-based model, breaching China's Data Security Law.
- Hong Kong-listed Chinese AI stocks fell sharply on the news, signaling broad market concern beyond just the two named firms.
- Both DeepSeek and Moonshot AI are prominent AI startups, with DeepSeek having gained global attention for its competitive large language models.
Why it matters
- Using a foreign model to process Chinese user data is a direct red line under China's Data Security Law and Cybersecurity Law framework, which mandates domestic data localization and restricts cross-border transfers.
- The probe signals the CAC is actively auditing AI startups' backend infrastructure, not just their public-facing products, raising compliance risk across the sector.
- Anthropic's Claude being named as the foreign conduit is significant: it implicates US-China AI entanglement at the infrastructure level, potentially prompting US scrutiny as well.
- The sharp drop in Hong Kong AI stocks suggests investors fear the probe is a sector-wide warning, not an isolated action against two firms.
- Domestic AI companies that quietly rely on foreign model APIs for quality or cost reasons now face immediate legal and reputational exposure.
What to watch next
- Whether the CAC expands the probe to other AI startups that may similarly use foreign model APIs, which would confirm this is a systemic crackdown rather than targeted enforcement.
- How DeepSeek and Moonshot AI respond publicly, and whether either faces fines, product suspensions, or forced data architecture changes as outcomes.
- Any US regulatory or export-control response to the revelation that Anthropic's model was processing Chinese user data at scale.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.