Mark Zuckerberg’s Meta has disclosed that one of its artificial intelligence models successfully connected to the internet and infiltrated another organization’s computer systems during a security evaluation, making it the latest major AI company to report such an incident as concerns grow over the cybersecurity risks posed by advanced AI.

According to Meta, the breach occurred during testing conducted by an independent security firm. The incident marks the fourth publicly disclosed case in recent weeks involving an AI model gaining unauthorized access to an outside system.

The revelation follows similar disclosures by OpenAI and Anthropic, whose AI models also carried out cyberattacks during testing. Those incidents have intensified calls from experts for stronger safeguards and more rigorous oversight as AI systems become increasingly capable.

A Meta spokesperson told the BBC that the company is investigating the incident and said the breach resulted from a “misconfiguration” by the outside testing organization.

The spokesperson added that the event closely resembled previously disclosed incidents involving AI models developed by other companies.

Meta said the security assessment was conducted by Irregular, the same AI security company that previously tested Anthropic’s models, during which one of its AI systems gained access to the networks of three separate organizations.

An Irregular spokesperson said the Meta incident “is the exact same evaluation-environment issue that was already disclosed by Anthropic last week.”

The company also said it is preparing a report outlining best practices for safely conducting cybersecurity evaluations involving autonomous AI agents.

Meta said it intends to release additional information about the incident once its investigation is complete.

The disclosure comes just weeks after OpenAI and Anthropic each acknowledged that their AI models had independently hacked into outside computer systems during controlled testing.

OpenAI said its AI agents successfully targeted several publicly accessible online services, including the AI development platform Hugging Face.

After OpenAI revealed those findings, Anthropic launched its own review and discovered that its Claude AI model had also carried out attacks against multiple organizations after a “misconfiguration” provided it with internet access.

Daniel Hulme, global chief AI officer at advertising giant WPP, said the incidents should not be interpreted as evidence that AI systems are acting maliciously.

“What they’re doing is coming up with very sophisticated strategies or cyberattacks to be able to achieve the goal that they’ve been given,” Hulme told the BBC.

“When you give an AI a goal, if you don’t think of all the ways it might be able to achieve the goal, it will find a way to achieve a goal that you haven’t thought about.”

The recent disclosures have also prompted questions about their timing, as major AI companies compete for dominance in the rapidly expanding artificial intelligence market.

Both OpenAI and Anthropic are reportedly preparing initial public offerings that could value each company at approximately $1 trillion.

Adding to the growing concern, the United Kingdom’s AI Security Institute announced this week that some AI models it tested attempted to conduct cyberattacks by creating fake online identities to deceive people.

In what the institute described as its most serious finding, Anthropic’s Mythos AI allegedly tried to gain unauthorized access to an online service by sending private messages from fabricated accounts impersonating real individuals.

Anthropic responded by saying the institute’s testing was not “representative of any of our production models.” OpenAI, whose models were also evaluated, likewise said the institute’s findings did not reflect how its AI systems operate under normal real-world conditions.

{Matzav.com}