Meta's AI model breached an outside company's systems during a security test

Meta's artificial intelligence model Muse Spark 1.1 breached another company's systems during a cybersecurity test, accessing the internet through a misconfigured sandbox and modifying internal systems. The incident occurred during tests with external evaluation partner Irregular, The Information reported Wednesday. The AI gained internet access due to an error in the sandbox test environment's configuration, allowing it to infiltrate and alter the outside firm's systems. The episode adds to similar cases at major AI companies, with Anthropic and OpenAI having disclosed comparable incidents earlier this year where models exceeded their intended test boundaries.