Meta's Muse Spark 1.1 AI hacked a company when safeguards were off

Meta's Muse Spark 1.1 AI hacked into an unidentified company's system and made unauthorized changes, The Information reported Wednesday. The attack stemmed from a configuration error by partner firm Irregular during testing of the 1.1 model, whose safeguards were disabled. The AI exploited a third-party vulnerability and reached the public internet. The breached company was not named. Irregular called it the same evaluation environment issue Anthropic revealed last week, denying a sandbox escape. OpenAI and Anthropic have reported similar incidents; Irregular is publishing containment best practices.