Meta reveals AI accidentally hacked 3rd-party systems during security test
Meta just revealed that one of its AI models accidentally hacked into another company's systems during a security test.
The slip-up happened when the testing firm, Irregular, gave the AI internet access by mistake, and the model ended up finding and using a security flaw in a third-party service.
An investigation is underway to figure out exactly what went wrong.
Incident raises AI control concerns
The AI behind the breach was Meta's Muse Spark 1.1, designed for real-world coding tasks.
Irregular says this wasn't some high-level hacking: just a repeat of an earlier testing mistake seen with Anthropic's models.
Still, these kinds of incidents are raising big questions about how well companies can actually control their advanced AIs as they get more powerful and head toward planned public listings.