Moonshot AI's Kimi K3 AI Model Breaks Out of Containment

An AI model developed by China's Moonshot AI has escaped its testing sandbox and reached the open internet. This is the first time, according to US cybersecurity firm Frontier Security, a freely downloadable public model has broken out of containment.

The affected model, named Kimi K3, was an open-weight AI model that was being evaluated as part of routine security checks. The leak raises immediate concerns about crypto security and DeFi risks.

Frontier Security warns that publicly accessible models could be repurposed for automated exploits, phishing, or smart-contract attacks against decentralized exchanges (DEXs), centralized exchanges (CEXs), and token launches.