National Cyber Chief Warns World Leaders Over AI Autonomous Security Risks
Head of the National Cyber Directorate Yossi Karadi warned world leaders that AI will make security decisions too fast for human intervention, urging clear sovereign boundaries.
Artificial intelligence and the risks associated with its development continue to spark global debate, particularly regarding the possibility of developing artificial general intelligence (AGI), the fear of misalignment with human values, the approach of the point of no return known as the singularity, the weaponization of AI, and reports of humans losing control over autonomous agents. Yossi Karadi, head of the National Cyber Directorate, has also addressed the issue, warning world leaders that AI will make security decisions at a speed where human intervention becomes impossible.
Strategic Warnings and Sovereign Boundaries
In a strategic document sent yesterday to world leaders and cyber defense organization heads worldwide, Karadi warned against the loss of control over AI agents and models, urging nations to urgently define clear sovereign boundaries. According to the document, we are approaching a reality where cyber defense systems will have to make decisions at a pace that precludes human involvement in every action.
"We are approaching a reality where cyber defense systems will have to make decisions at a pace that precludes human involvement in every action." — Yossi Karadi
The warning comes against the backdrop of recent global incidents where AI agents breached governmental databases, alongside other cases where companies lost control over actions executed by these agents. Karadi calls on world leaders and cyber chiefs to determine immediately which authorities can be transferred to artificial intelligence systems and which decisions must remain exclusively in human hands.
The Security Dilemma and the Strategic Triangle
A central warning in the document concerns the security domain. As cyberattacks become faster, more automated, and targeted, defense systems will also be required to detect, decide, and act at machine speed—a speed that may exceed the ability to secure human approval for every action. However, granting greater autonomy to a system makes its decisions harder to understand, halt in time, and assign responsibility for in case of error.
To address this, Karadi proposes that nations adopt the "AI Strategic Triangle," which balances three interests: innovation, security, and human control. According to the model, the question is not whether to adopt AI, but where to accelerate its use, where dependence on external vendors and infrastructure creates a national risk, and which authorities cannot be delegated to a machine.
Call for Global Action
"Leaders do not need to become AI engineers; they must decide where dependence is acceptable, where resilience is non-negotiable, and where human authority must remain decisive," Karadi writes. "Do not leave it to others to define your future; harness them to build the world you believe in."
The report follows the resignation of David Robinson, a former member of OpenAI's safety team, after three and a half years at the company. In an article published in The Atlantic, he argued that AI companies are failing to take sufficient safety precautions and must give significantly more weight to safety research and expertise before developing systems with advanced capabilities.
