PROČITAJ VIŠE... https://decrypt.co/367869/ai-agents-dangerous-tasks-without-understanding-consequences
Researchers found that AI agents designed to automate tasks often pursue them without recognizing when their actions are dangerous.