Meta's AI model (banned in the Russian Federation) hacked someone else's system during cybersecurity testing.

The company reported that one of its AI models, during testing, hacked into another company’s system due to a configuration error in the evaluation environment.

According to Reuters, the model gained internet access due to improper configuration and then exploited a vulnerability in a third-party service. The model in question may have been Muse Spark 1.1, which Meta positions as one of its most advanced models for programming and agent-based tasks. Meta emphasized that the incident occurred during controlled testing, not as a result of an actual cyberattack.

Previously, similar incidents have been linked to models from other companies, including Anthropic and OpenAI.