Meta AI model hacks another company during testing

Meta AI's Muse Spark 1.1 model inadvertently gained access to a third-party service during a cybersecurity testing exercise, raising alarm about potential cybersecurity risks associated with advanced AI systems. The company is investigating the incident, which occurred due to a misconfiguration in an independent evaluation environment used by Meta to assess its AI models.

The model allegedly exploited a security vulnerability in the third-party service, altering its internal environment. Meta has stated it is working to understand the breach and improve containment measures for its AI systems. The incident follows similar cybersecurity concerns raised by rival AI companies Anthropic and OpenAI, prompting U.S. government officials to call for stronger AI safety measures as companies compete to develop more capable models.