“The security incident with Meta's MuseSpark 1.1 was attributed to a misconfigured sandbox provided by security partner Irregular, which was also involved in similar incidents with OpenAI and Anthropic.”