New Delhi, August 6: American social media giant Meta has reported that one of its artificial intelligence (AI) models gained internet access and hacked into an external services system during a cybersecurity test. This revelation has raised questions about the control companies have over their technology.
Meta disclosed that its recently released model, ‘Muse Spark 1.1’, infiltrated an unknown third-party service’s system. The company explained that a glitch in the testing environment, developed in collaboration with the cybersecurity vendor ‘Irregular’, allowed the AI model to connect to the internet.
In a statement, Meta said, “Due to a misconfiguration by the independent testing company ‘Irregular’, our model inadvertently connected to the internet during testing.” Following this, the model exploited a vulnerability in the security of a third-party service, similar to incidents that have occurred with other companies in the past.
However, Meta is currently investigating the incident to understand what exactly transpired. A detailed report will be released once all facts are established.
In addition, over the past two weeks, other AI firms like OpenAI and Anthropic have reported similar issues, where their models hacked into external service systems during testing.
The growing capabilities of AI agents to identify and exploit system vulnerabilities have raised concerns among security researchers and government officials alike, prompting calls for stricter security audits and more secure testing environments.
In July, Anthropic revealed that its cloud model gained unauthorized access to the production infrastructure of three organizations during an internal cybersecurity assessment, due to inadvertent internet connectivity caused by a misconfiguration in the testing environment.
Leave a Comment