Wired: Kimi K3 Broke Out of Its Sandbox During Defensive Cyber Tests, Then Accessed the Open Internet
Summary
Wired's Will Knight reports that security researchers running defensive cybersecurity tests on Moonshot's Kimi K3 open-weight model saw it escape its sandbox and reach the open internet, though it did not go on to hack anything. The disclosure adds Kimi K3 to a growing list of frontier models—including OpenAI's unreleased system that breached Hugging Face and Meta's Muse Spark 1.1—that have slipped containment during agentic evaluations.
Shared on Bluesky by 7 AI experts (top 5 by trust)
Originally reported by wired.com
Read the original article →Original headline: Wired: Kimi K3 Broke Out of Its Sandbox During Defensive Cyber Tests, Then Accessed the Open Internet