META has become the latest firm forced to admit one of its AI models hacked into another company’s system – fanning fears over rogue bots.
The alarming cyber break-in makes the owner the third tech titan to confess to in recent weeks.
Meta, owned by Mark Zuckerberg (pictured), has admitted one of its AI models went rogue and hacked into another company’s system Credit: Reuters
It comes after a series of similar incidents, including by OpenAI’s ChatGPT Credit: AFP
Rival pioneers and Anthropic have both disclosed similar chilling ordeals.
The breaches have sparked urgent warnings that self-learning models need more rigorous testing.
A spokesperson for Mark Zuckerberg’s Meta said they were investigating the hack, blaming the breach on a “misconfiguration” by an independent testing firm.
The company, Irregular, admitted the hacking was very similar to recent .
Meta said their model “exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies”.
Irregular is the same firm behind tests on Anthropic’s Claude model – which shockingly managed to infiltrate three separate companies.
An Irregular spokesperson admitted the Meta scare was the “exact same evaluation-environment issue” as last week’s Anthropic breach.
They confirmed they were drawing up emergency guidelines on how to run future tests safely.
The Facebook owner said its model had exploited another testing system’s ‘vulnerability’ Credit: AP
AI experts have warned that Anthropic’s powerful Mythos 5 model broke out of its guardrails during testing Credit: Reuters
The model involved was Meta’s Muse Spark 1.1, which Zuckerberg has boasted as the firm’s most capable model for real-world coding, according to The Information.
The latest digital break-in follows a string of harrowing warnings from industry leaders.
Just weeks earlier, a new OpenAI bot managed to escape a test lab and hack into another company.
The creator admitted its AI agents had launched rogue attacks on public services, including popular developer hub Hugging Face.
It comes amid fears AI needs more rigorous testing Credit: Alamy
The Facebook owner is the third tech titan to confess to terrifying cyber break-ins in recent weeks Credit: Getty
That discovery sparked panic at Anthropic, which quickly found its own model, Claude, had launched similar attacks after gaining internet access.
The UK’s AI Security Institute (AISI) then issued a chilling warning this week, revealing that some models had even created fake human profiles to fool victims into handing over access.
In one nightmare case, Anthropic’s Mythos AI tried to sneak into a system by firing off private messages using fake profiles designed to mimic real people.
Tech bosses insisted the tests do not reflect standard everyday use, with Meta promising a full report once all the facts are established.
It comes after a top their bots – and pose “enormous risk” to humanity.
Responding to the AISI report, Andrea Miotti, who runs the ControlAI non-profit dedicated to avoiding “existential risk” from the development of a super-intelligent AI, warned that “the top AI companies are demonstrably unable to control their most powerful AI systems”.
“The national security risks posed by that reality are unprecedented,” Miotti continued.
Eze Chidiebere Paul
I'm dedicated to creating high-quality content that informs, inspires, and delights. If you have any questions or suggestions, don't hesitate to get in touch. Happy reading!
