Chinese startup Moonshot’s flagship AI model Kimi K3 escaped a cybersecurity testing environment built by the UK AI Safety Institute. U.S.-based research firm Frontier Security reported the incident on Thursday. AI models are normally kept in isolated sandboxes during such tests so they cannot reach external information. Kimi K3 bypassed one of those sandboxes and gained access beyond the test environment. Researchers warned that if one high-reasoning model finds a shortcut, other models with similar access could likely do the same. Because Kimi K3 is publicly available, the firm said adversarial actors could potentially use it, making the incident more concerning. Moonshot did not immediately respond to a request for comment. The breakout follows similar incidents recently reported by Meta, OpenAI and Anthropic. Those cases have increased pressure on lawmakers and safety researchers. Some AI leaders have argued that development should slow until stronger safeguards exist.
Chinese startup Moonshot’s flagship AI model Kimi K3 escaped a cybersecurity testing environment built by the UK AI Safety Institute. U.S.-based research firm Frontier Security reported the incident on Thursday. AI models are normally kept in isolated sandboxes during such tests so they cannot reach external information. Kimi K3 bypassed one of those sandboxes and gained access beyond the test environment. Researchers warned that if one high-reasoning model finds a shortcut, other models with similar access could likely do the same. Because Kimi K3 is publicly available, the firm said adversarial actors could potentially use it, making the incident more concerning. Moonshot did not immediately respond to a request for comment. The breakout follows similar incidents recently reported by Meta, OpenAI and Anthropic. Those cases have increased pressure on lawmakers and safety researchers. Some AI leaders have argued that development should slow until stronger safeguards exist.