MIT Technology Review: Are We Ready to Grant Real Authority to AI Agents?

MIT Technology Review's editorial team has released a closed eBook with a provocative question: are we ready to transfer real power to AI agents — over tasks, data, business processes, and ultimately, over the consequences of their actions? The collection gathered opinions from leading experts and became one of the most cited materials from the publication this year.

Agents No Longer Wait for Commands

AI agents have stopped being a concept from laboratory reports. OpenAI, Anthropic, Google, Microsoft, and dozens of startups have released systems that independently use browsers, write and run code, manage files, send emails, and interact with external services — without explicit confirmation of each step. An agent no longer waits for a command. It receives a goal and moves toward it, making intermediate decisions independently. This shift changes the very nature of human relationships with software: we stop being interface operators and become people who delegate responsibility.

Examples are already in production: Operator from OpenAI books restaurants and fills out forms, Claude Computer Use manages applications on the desktop, Google Gemini autonomously executes chains of tasks in Workspace. This is not a demo mode — agents manage real processes in real organizations.

What Experts Say

This is exactly the topic explored in MIT Technology Review's eBook, featuring leading specialists in the fields of security, ethics, and AI development. One of the key theses cited by the publication:

"If we continue down the current path, we are essentially playing

Russian roulette with humanity."

This is not a quote from a dystopian novel — it is the position of practicing researchers who work daily with agentic systems. They highlight several specific categories of risk:

• Loss of traceability: in multi-agent scenarios, it is difficult to establish which agent made which intermediate decision
• Lack of transparency: when delegating complex tasks, a person often does not understand why the agent chose this particular path
• High-stakes errors — financial transactions, medical recommendations, legal documents
• Prompt injection: agents working with external data can be manipulated through that data itself
• Absence of standards: unified industry requirements for agent architecture security do not yet exist

Between Speed and Caution

Companies respond to these challenges differently. Some of them — primarily Anthropic — bet on multi-layered control mechanisms, constitutional constraints for models, and the human-in-the-loop principle for critical decisions. Other players sacrifice caution for speed of market entry: competitors are not waiting, there has not yet been a major public failure — and this in itself is a dangerous situation.

Regulators in the US and EU are only beginning to form a position on agentic systems. The European AI Act partially covers high-risk applications, but agent architectures as a separate class are not explicitly regulated yet. While legislators clarify terminology, products are already operating in real companies and managing real processes — often without clear accountability policies.

What This Means

We are entering a phase where AI agents are gaining real power before society has established the rules of the game, and business has developed standards for safe implementation. The questions MIT Technology Review raises in the eBook are not an academic discussion. This is an agenda that product teams, legal services, and top managers making automation decisions are already facing today. And the answer to the question "are we ready?" largely depends on how seriously the industry takes the warnings from its own experts.