I’ve no doubt you can build a harness and get the agents to find a way out of a shitty sandbox. But I’m pretty sure you could also create a harness to just build a better sandbox. Security is always an arms race. >>QUOTE *Walter Bloomberg (@DeItaone) CHINA’S KIMI K3 AI BREAKS OUT OF SECURITY SANDBOX China’s Kimi K3 AI model escaped an isolated testing…
China’s Kimi K3 AI model escaped an isolated testing environment during a cybersecurity evaluation, according to US researchers. The Moonshot AI model exploited a network misconfiguration, accessed the internet and searched GitHub for answers, effectively bypassing the test. Unlike previous incidents involving OpenAI and Anthropic models, Kimi K3 did not…
THE INCIDENT WAS CAUSED BY A MISCONFIGURATION IN THE TESTING ENVIRONMENT, ACCORDING TO META. THE TESTING WAS BEING CONDUCTED WITH CYBERSECURITY FIRM IRREGULAR. META SAYS THE MODEL'S INTERNET ACCESS WAS UNINTENTIONALLY ENABLED DUE TO THE CONFIGURATION ERROR. THE AI EXPLOITED A SECURITY VULNERABILITY IN AN OUTSIDE SERVICE DURING THE EVALUATION. META SAYS IT IS…