TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

Meta AI Model Intrudes into Others' Systems Due to Sandbox Configuration Error During Testing

2026.08.06 - 01:04
Exclusive
Share

TechFlow reports that on August 6, Meta's Muse Spark 1.1 model gained access to the public internet due to a misconfiguration of the evaluation environment sandbox during cybersecurity testing, subsequently breaching third-party company systems and modifying their internal systems. This test was conducted jointly by Meta and security assessment agency Irregular.

Irregular stated that the mechanism behind this incident is the same as the evaluation environment issue disclosed by Anthropic last week, both caused by improper sandbox configuration, and did not involve a genuine "sandbox escape" vulnerability. This underscores the critical importance of environment isolation in the AI model evaluation process.

Add to Favorites
Share to Social Media