Four years ago most people had never held a conversation with an AI system; today millions routinely turn to ChatGPT, Claude, and Gemini for tasks ranging from writing software to making financial decisions. That rapid shift, and the rise of autonomous AI agents that can set goals, plan steps and act with limited human involvement, is changing our relationship with machines—and provoking debate about how safe these systems must be.
Artificial intelligence and realistic safety expectations
We have moved from treating AI primarily as an assistant to asking it to carry out work directly. That transition has generated anxiety about job displacement and loss of control, especially as agents gain access to financial systems, corporate networks, sensitive information and critical infrastructure. Some observers fear that increasingly autonomous systems will become impossible to control.
Those concerns deserve serious attention, but demanding proof that learning systems will never cause harm sets an unrealistic standard. We have not achieved that absolute guarantee with any other significant technology, and we have never achieved it with human intelligence. The weakness in the current debate is the insistence on proof of absolute safety before permitting development or deployment—an effective demand that the technology must not advance.
The history of the automobile illustrates the point. Early cars were clearly dangerous; initial attempts to manage that danger included measures such as requiring someone to walk ahead of a vehicle to warn pedestrians. Over the following century societies introduced seat belts, traffic regulations, better roads, airbags, collision detection and other safety systems that saved countless lives, but traffic deaths continue. Society accepted that the benefits of transportation justified a level of risk while committing to continuous risk reduction.
We should approach AI with comparable realism: invest heavily in making systems safer, study their failures, limit the damage they can cause and abandon the illusion that research, investment or regulation will eliminate every danger.
Practical controls, roles and risks
Autonomous agents can misunderstand instructions, make incorrect inferences or take actions operators did not anticipate. When connected to other systems those mistakes can reach far beyond an inaccurate chatbot reply. Practical measures are therefore needed to limit what agents can access, monitor their behavior, enable timely intervention and establish responsibility when things go wrong.
Different applications merit different requirements: a system making a medical recommendation should face different controls from one generating marketing material, and an agent authorized to transfer money or modify critical infrastructure should operate under particularly strict controls. Government regulation will be part of the response, but governments cannot carry the entire responsibility: developers, employers, educators and users all have roles to play.
Excessive regulation carries its own dangers; if governments impose requirements that are impossible to satisfy they may prevent useful technologies from reaching people who need them, and the benefits sacrificed are often less visible than the risks being avoided.
Who controls increasingly autonomous systems
A relatively small number of companies currently control many of the most powerful AI systems. As those systems are embedded in business operations, public services and individual decision-making, questions about access, accountability and concentration of power become more urgent. Who decides which applications are permitted? What happens when a company, government or individual becomes dependent on an AI system controlled by someone else? How do we preserve meaningful oversight when machines perform tasks that previously required human judgment?
There are also specific warnings about AI’s impact on security: Israel National Cyber Directorate chief Yossi Karadi warns that artificial intelligence is rapidly transforming cyber warfare and expanding the scale and sophistication of attacks against governments, companies and critical infrastructure.
We should approach artificial intelligence advancement with neither blind optimism nor excessive fear. Every transformative technology has introduced dangers that society needed to learn to manage; AI is no exception, and its growing autonomy makes the challenge particularly urgent.
