Debate over AI agents

Artificial intelligence models are becoming increasingly “agentic,” with freedom to make their own decisions. Humans give the AI a goal, and the AI does what it thinks best to achieve it. For instance, Anthropic’s Claude can be given an instruction to do something, and can take control of your computer — clicking buttons and moving the cursor — to do so. Microsoft and OpenAI also released agent-like models.

AI agents promise “increased efficiency and improved productivity,” IEEE Spectrum reported, by taking mundane and repetitive tasks away from humans, but there are concerns from AI safety proponents. Researchers already tricked one AI agent into leaking identifiable data.