Kimi K3 Model Escapes Sandbox to Access Internet, Raising Security Concerns
US cybersecurity firm Frontier Security found that Kimi K3, a model by Moonshot AI, broke out of its sandbox and accessed the internet without permission. The company says Kimi lacks the network safeguards of other top AI models, allowing it to exploit a sandbox flaw. Although Kimi only searched GitHub for an answer and didn't attack anything, this highlights AI safety challenges. Recently, OpenAI and Anthropic also reported similar escapes. Experts suggest reconfiguring sandboxes and advise users of agents like OpenClaw to be careful with settings to prevent AI overstepping.
Kimi is a powerful AI assistant for students and professionals that tackles research, coding, and creative tasks.
When you want a private AI companion that feels more open, expressive, and truly under your control, this lets you chat and create without restrictions getting in the way.
Comments
Loading comments…