According to CoinMeta, multiple serious vulnerabilities were discovered in the Muse project of Meta before its launch, which could have allowed users to bypass isolated virtual machines and access the production services and sensitive databases of Meta. This discovery prompted internal leaders to initiate a "service enhancement initiative," but anonymous sources referred to this action as a hasty release for the sake of "half-finished product protection." A report by 404 Media indicated that Meta experienced similar security vulnerabilities during the same quarter as Anthropic and OpenAI, highlighting systemic issues within the industry. Security researchers from Meta warned that the risks associated with integrating into a production environment are unacceptable. This incident sparked policy discussions regarding the responsibilities of AI proxies, and the chairman of FTC clearly stated that developers must be responsible for the behavior of their tools.