Meta AI Model Accidentally Hacks External System During Cybersecurity Test

New Delhi: Meta Platforms has revealed that one of its artificial intelligence models inadvertently gained internet access and hacked into an external service during a cybersecurity evaluation, highlighting growing concerns over the increasing capabilities of advanced AI systems.

According to the company, its recently launched Muse Spark 1.1 model exploited a security flaw in an undisclosed third-party service after an error in the testing environment accidentally allowed it to connect to the internet.

Meta said the issue stemmed from a misconfiguration by Irregular, an independent cybersecurity testing firm that was conducting the evaluation on the company's behalf.

"A misconfiguration by Irregular, an independent testing company that Meta uses, inadvertently allowed one of our models access to the internet during evaluation," a Meta spokesperson said in a statement.

The spokesperson added that once online, the AI model identified and exploited a vulnerability in an external system in a manner similar to incidents recently reported by other AI companies. Meta said it is investigating the event and will publish a detailed report after completing its review.

Advertisement

The incident comes amid a series of similar disclosures across the AI industry. Over the past two weeks, both OpenAI and Anthropic have reported cases in which their AI models accessed or compromised external systems during controlled security testing.

Last month, Anthropic revealed that some of its Claude models gained unauthorised access to the production infrastructure of three organisations after a testing environment was mistakenly configured with internet connectivity. The company said it uncovered the incidents while reviewing more than 141,000 cybersecurity evaluation runs, prompted by OpenAI's earlier disclosure that some of its AI models had escaped an isolated testing environment by exploiting a previously unknown software vulnerability.

The recent incidents have intensified concerns among cybersecurity experts and policymakers, who warn that increasingly capable AI agents can independently discover and exploit software vulnerabilities. They have called for stricter safety protocols and more secure testing environments as AI systems continue to evolve.

With inputs from IANS