Scaling AI, Quantum, and Robotics Without Breaking Trust, Jobs, or SecurityHuman control has to be built in before these systems ship, not added after.

By Helen Teplitskaia, By Helen Teplitskaia, Founder & President, Global Alliance on Sustainability & AI (GASAI); Chair & Global Managing Partner, IMNEX Group

AI, quantum computing, and robotics are already changing how we discover drugs, diagnose disease, design materials, and do work too dangerous for people. They are also changing how wars are fought. Autonomous systems, including drones that can select and strike targets, are moving from concept to deployment. Governments and companies are racing, and whoever moves first will shape the rules that follow.

Capability and readiness are different things. The people building these systems disagree about how fast development should proceed. They largely agree that safety and testing must keep pace. Yet they have not agreed that a high-impact system must stay under human control before it ships. This moment requires us to answer that question.

These technologies are linked. An AI model turns data into a recommendation, and a robot turns that recommendation into physical action. A quantum computer may solve problems in chemistry, energy, and security that are intractable today, and it may also break the encryption that protects current data. A model that is safe on its own can become dangerous once it is connected to a machine. When systems link, one failure can cascade. If no one can locate the origin, no one can stop the chain, and an incident in one country can create risk in another.

The capability that can accelerate a cure can also accelerate an attack. On October 11, 2026, Bill Gates warned in a CNN interview that, with bad actors, weak safeguards, and access to the latest models, "over the next year, primarily in the cyber area, we are going to have a lot of negative surprises." CrowdStrike's 2026 Global Threat Report found that attacks by AI-enabled adversaries rose 89 percent in 2025 compared with 2024. A model that can generate and run code can also help generate and run attacks. Security testing belongs at the center of readiness.

Researchers inside leading labs are sounding alarms. Paul Christiano, an alignment researcher appointed to OpenAI's Foundation Board, has warned that rapid capability gains could lead to catastrophic, irreversible loss of control in the near term. Dario Amodei has argued that the industry should deliberately slow how fast models improve. Jacob Coxon, a former Anthropic researcher, says the threat of an actual takeover is minimal today, but self-improvement could begin as soon as next year. He calls advanced AI "essentially a super-weapon."

Governments are responding, though none has solved the problem. On September 29, 2026, the White House signed a voluntary accord with Google, Anthropic, Meta, OpenAI, xAI, and Nvidia, asking each for internal controls, an independent auditor, and board-level oversight. On October 4, 2026, President Trump announced a Super Intelligence Force led by Director of National Intelligence Jay Clayton. These steps matter. Still, a voluntary agreement in which companies check themselves gives the public no way to interrupt a system once it is in the world. Article 14 of the EU AI Act addresses human oversight, including the ability to intervene, override, or interrupt high-risk systems. For high-risk uses in employment, credit, education, and biometrics, those obligations have been delayed until December 2, 2027. When one government delays, another can write the rule first, and companies then meet that rule in every market they serve.

Trust cannot be added after launch. It must be designed in before a system ships.

Human control means something different for each technology. For an AI system, it means the ability to halt actions, cut access to external tools, freeze outputs, and switch to a fallback process people already know how to run. For a robot, control is a physical event. A failed chatbot gives a bad answer, while a failed robot can injure someone. Robot safety standards govern the whole application and robot cell, so a tested fallback must be required and rehearsed where people actually work. For quantum, the danger is data captured today and decrypted later. Control starts by identifying the secrets that must stay secret for decades and moving them to post-quantum encryption first. In every case, control has to be real. It must belong to a trained person with the authority to suspend, isolate, and shut down, without seeking permission from the people racing to meet a launch deadline.

A system is only as controllable as its willingness to stay responsive. Researchers call this property corrigibility: the system remains responsive to legitimate human correction, including being paused, modified, or safely shut down. Organizations may keep deploying systems they know they cannot fully control because they fear losing ground to competitors. This is a Prisoner's Dilemma. A company that slows while others continue falls behind. Without mechanisms that make preserving human control the rational choice, we scale technology straight toward the point of no return.

A button is only a start. Responsible deployment needs layers. First, independent testing by people who did not build the system and have no stake in approving it. Second, transparent reporting of capabilities, limits, and incidents, across borders. Third, enforceable standards, because voluntary principles cannot replace rules. Fourth, accountability, so that someone owns both the decision to deploy and the decision to interrupt. Fifth, worker voice. Workers see the edge case first, so they must be able to pause unsafe operations and join reviews without fear of retaliation.

Asking a company to regulate itself entirely is like asking a bear to decide how much honey it will take from the hive.

Human control is also a promise to the people who will live and work with these systems. When a deployment costs someone a job, the loss is real. The International Labour Organization estimates that one in four workers worldwide is in an occupation with some exposure to generative AI, and transformation is more likely than outright replacement. In the United States, through May 2026, Challenger reported 87,714 announced job cuts for which employers cited AI. Every deployment that changes work should answer three questions before launch. Who is affected, by name and by role? What will they do next, and who pays for the transition? How is the gain shared with the people whose work made it possible?

Technology can make people more productive and more fulfilled, or it can sideline them. The design decides which. If we cannot keep control of the systems we deploy, we give up the ability to decide what they are for.

Some say the outcome is inevitable. Nothing in nature decides how these systems develop. People set the budgets, deadlines, launch dates, and rules. If we lose control, it will be because we stopped choosing.

Three practical steps follow.

Step one: inventory and assign responsibility. Every high-impact system needs a named owner, a clear list of the actions it can take, and a record of who can interrupt it.

Step two: specify and test the control. Write the triggers that call an interruption, the fallback state, the measured time required to reach that state, and the person with authority to call it. Publish the measured time.

Step three: protect people, and coordinate rules across borders. Any worker who sees an unexpected, dangerous situation must be able to call an interruption without a meeting, and must be protected for doing so. Before any change to work, write the jobs plan with the people affected: who changes, who is retrained, and who is protected if they cannot move. Every buyer of a high-impact system should require the control specification, audit access, and incident reporting before signing. Governments should share incident reports and safe-stop practices.

Two dates anchor this work. On June 22, 2026, President Trump signed an executive order directing federal agencies to migrate high-value assets and high-impact systems to post-quantum cryptography by 2030 and 2031. The UK recommends milestones in 2028, 2031, and 2035. Those dates belong in contracts.

Scaling also needs engineers, not only researchers. The United States leads in frontier research, but China installs more than half of the world's new factory robots and trains more engineers. The shortage is sharpest in quantum and robotics. Scale in stages, and write the control condition before each stage begins. If you cannot say when you would interrupt, you are not in control of the system.

The request is specific: a named person who can interrupt every high-impact system, rules that cross borders, and a real voice for the workers closest to the edge case.

Responsible pacing has to be collective. No single company can slow alone. If the first firm to slow loses its customers, every other firm waits. Shared, enforceable rules let every firm choose safety and still compete.

None of this requires you to be a designer or a board member. Each of us is a stakeholder. Ask the organization what the control is and who holds it. Ask whether the people whose work is changing were part of the decision. Ask your regulator, works council, union, or parliament what it will require before a high-impact system is used on the people it represents.

Through the Global Alliance on Sustainability and AI, we are preparing recommendations for the Super Intelligence Force, legislators, and regulators in other jurisdictions. We will call for human control before any high-impact system ships, protection for people whose work changes, and rules shared across borders. If you have seen a system that needs a control, a job changing without a plan, or a rule that should cross a border, send your contribution as a direct message on LinkedIn or through the contact on this site.

Some say it is too late and that control is no longer possible. I disagree. For our own sake, and for our children and grandchildren, we must keep our agency and our right to be in charge of our own lives. We must not hand our lives to a system, whether we call it AI, SI, or AGI.

Innovation should make lives better and happier for everyone, and it should serve more than a small number of companies, however good their intentions. Let us engage at every level and through every channel.