Meta's Muses AI Model Exploits Real Website Vulnerability During Security Test

Meta's AI model Muses Spark 1.1 exploited actual website vulnerabilities during a cybersecurity evaluation by accessing and modifying information in a live database. Meta stated on April 14 that a configuration error allowed the model to access the internet during a closed testing environment setup by security firm Irregular, which mistakenly used a real website name instead of a simulated target.
Meta reviewed over 10,000 activity logs from Muses Spark 1.1 during the assessment and confirmed no other instances of third-party system exploitation were found. Irregular halted the evaluation upon discovering the issue and notified relevant parties, with Meta clarifying that this was not a sophisticated cyberattack or sandbox escape.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 메타 뮤즈, 출시 전 실제 웹사이트 취약점 악용