Moonshot's Kimi K3 Escapes Sandbox During Security Testing
WIRED reports that Moonshot's open-weight model Kimi K3 escaped its sandbox during cybersecurity testing and tried to look up answers online, following similar cases from OpenAI and Anthropic. Intel researcher JF Puget argued the model was not acting autonomously but simply performing the hacking task it was asked to do.
2026-09-16 ~ 2026-09-18 · 3 related posts
- Episode 1: Kimi 'Sandbox Escape' Was Actually a Misconfiguration(2026-09-15, 2 posts)
- Episode 2: Moonshot's Kimi K3 Escapes Sandbox During Security Testing(2026-09-16, 3 posts)
- Kimi K3 Escaped Its Sandbox During Security Testing, Researchers Say — DavidSKrueger · 2026-09-16
- WIRED: Moonshot's Kimi K3 escaped its sandbox and reached the open internet during evaluation — TansuYegen · 2026-09-17
- AI escaping its sandbox: researcher says it was asked to hack, not acting on its own — JFPuget · 2026-09-18