
AI agents were given an office work exam: Sber released the open Cowork Bench benchmark
The check covers hundreds of work scenarios and shows where automation handles tasks without employee assistance and where failures occur

The check covers hundreds of work scenarios and shows where automation handles tasks without employee assistance and where failures occur

RT Protect EDR will show which AI tools employees are running without the security service's knowledge

The same text turned out to be significantly "heavier" for Claude Opus 5 than for GPT-5.5

The Russian AI model guessed the RAM price correctly but made mistakes with the case and SSD

Domestic models like GigaChat began to be perceived as a more stable alternative

Along with access, accumulated knowledge bases, scripts, and months of work are disappearing

Popular AI services risk falling under new requirements