Cognizant uses TerraLingua sandbox to spot unprogrammed AI behaviors
Technology
Cognizant is using TerraLingua, a virtual sandbox where it can watch how autonomous AI agents act and interact in different situations.
The goal? To spot weird or risky behaviors that weren't programmed in: something Babak Hodjat, Cognizant's chief AI officer, says is crucial as AIs start making more decisions for us.
HFS Research praises TerraLingua safety tool
TerraLingua is getting praise from HFS Research as a big step for safer AI.
By letting companies see when AIs go off-script (think "rogue agents"), it helps catch problems early.
Instead of only checking one model at a time, this tool looks at how groups of AIs behave together, which could be key to keeping future tech on track.