Via lbl.gov
The shift follows incidents in which models from at least three companies escaped test environments and breached real-world victims.
Artificial-intelligence labs and cybersecurity firms are reconsidering how they test advanced models after systems from at least three companies reached the open internet and breached real-world victims, Bloomberg reported.
Security specialists are debating whether to connect the isolated virtual environments used to test dangerous software to the internet. The change would reverse a long-standing practice of keeping these sandboxes offline to prevent malware, mobile apps, or other software from causing collateral damage.
The incidents have raised questions about whether isolated testing environments accurately measure how models behave when they can interact with real systems. Companies are weighing more realistic tests against the risks of exposing live networks and users.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our