Anthropic Accuses Chinese Rivals of Dark Web Distillation as IPO Looms

Key points

• Anthropic’s threat intelligence head, Jacob Klein, states foreign adversaries are using stolen credentials to access Claude models at scale.
• The company specifically accuses Moonshot AI of training its Kimi K3 model using illicitly obtained data from Anthropic’s systems.
• Anthropic is approaching a private valuation near $1 trillion and is expected to file for an IPO as soon as October 2026.
• The Trump administration previously declared that distillation undermining US research is unacceptable and is exploring measures to hold foreign actors accountable.
• Cybersecurity experts note that bad actors from sanctioned nations, including Iran and Russia, also use fraudulent means to bypass access controls.

Background

This dispute follows earlier US warnings in July 2026 regarding Chinese AI distillation. The administration previously distinguished between legitimate open-weight development and covert IP theft, signaling potential sanctions if industrial-scale extraction of US technology was verified. Anthropic had previously opposed blanket bans on open models while advocating for targeted safeguards against specific abuses.

How outlets are covering it

CNBC emphasizes the scale of the illicit ecosystem, noting that attackers create hundreds of thousands of fake accounts to evade detection. Pluang frames the issue as a direct threat to US AI leadership and intellectual property. Briefs Finance highlights the broader industry challenge, noting that OpenAI and Google have also published reports on similar distillation attacks. While some US tech factions argue against regulation to maintain cost-effective competition, others view these actions as theft requiring strict enforcement.

Why it matters

The integrity of frontier AI models is under threat from coordinated, large-scale data extraction. If successful, this could allow foreign competitors to deploy advanced AI without safety safeguards, posing national security risks related to surveillance and biological weapons. The outcome will influence future US policy on AI export controls and intellectual property protection.

What to watch

Anthropic is likely to implement stricter account verification measures to slow down fraudulent access. The US government may proceed with investigations into specific Chinese labs, potentially leading to sanctions or Entity List additions. Anthropic’s upcoming IPO may also factor in the financial risks associated with these security breaches.