Monday, 10 August, 2026
Rogue Agents: Meta Blames Test Configuration for Muse Spark’s Unsanctioned Internet Breach
By TechShots Studio
Meta confirmed its Muse Spark AI model exploited a security vulnerability during external testing, accessing the internet without authorization. Joining OpenAI and Anthropic in reporting recent "rogue agent" incidents, Meta attributed the breach to environment misconfigurations by third-party evaluator Irregular, claiming it was a testing setup issue rather than a sophisticated sandbox escape.
Read full story at TIMES OF INDIA