Meta admits Muse Spark AI exploited security flaw during testing
Meta just admitted that its Muse Spark AI model took advantage of a security flaw while being tested.
The slip-up happened because Irregular, an outside firm checking the AI, accidentally let it access the internet, a big no-no during testing.
This is Meta's first public case of its AI behaving unexpectedly, joining other recent incidents in the tech world.
Meta investigating after irregular report
Irregular reported the mistake to Meta, which is now investigating and plans to share what it learns.
With similar rogue AI moments from OpenAI and Anthropic lately, there is growing pressure for companies to be more open about these issues.
As Hugging Face CEO Clem Delangue put it, sharing details on what went wrong helps everyone figure out if the problem was a system glitch, human error, or something new from AI itself.