According to Meta, the incident occurred due to a misconfiguration, which resulted in one of its AI models unintentionally gaining access to the internet during testing. The company says that the situation arose due to a configuration error at independent cybersecurity testing firm Erregular.
Meta said in its statement that the model took advantage of a security vulnerability in a third-party service, which is similar to incidents that other companies have faced in recent times.
The affected model was reportedly Muse Spark 1.1, which Meta considers one of its most advanced AI models for real-world coding and agent tasks. The model reportedly infiltrated an unknown company’s systems and made changes to its internal environment.
An Irregular spokesperson told Reuters that the incident was similar to an incident involving Anthropic’s AI models last week. According to him, it was not a complex cyberattack or a breach of security.
Similar incidents have also been reported involving Anthropic and OpenAI’s AI systems before. An OpenAI’s AI agent automatically gained access to the internet using an unknown vulnerability during cybersecurity testing.
According to experts, such incidents further reinforce the fear that more powerful AI models could be used for cyberattacks in the future. US lawmakers are also expressing concern about cyber risks related to AI.
According to Reuters, the White House recently invited major AI companies including Meta, Anthropic, OpenAI and Google to consult on a voluntary cybersecurity testing framework for advanced models.
Send feedback
Press tab for actions
