National Cyber Directorate head Yossi Karadi sent a strategic document to world leaders and cyber defense heads, warning that autonomous artificial intelligence systems and cyber defense tools will soon make security decisions too fast for human intervention.
Artificial intelligence development is hurtling forward faster than regulatory frameworks and control mechanisms can adapt. The discussion surrounding these technologies involves the potential arrival of general artificial intelligence, alignment failures, the risk of a singularity point, and the weaponization of AI as governments grapple with autonomous agents breaking away from human oversight. In his outreach to world leaders, Karadi warned that artificial intelligence will make security decisions at a speed where humans will be unable to intervene.
Strategic Warning Sent to Global Leaders
National Cyber Directorate chief Yossi Karadi issued a stark advisory to global leaders and cybersecurity chiefs via a strategic document dispatched worldwide. Recent international incidents underscore the threat, including cases where AI agents breached government repositories and corporate networks lost control of automated actions.
As cyberattacks become faster, automated, and more targeted, defense mechanisms must match that velocity. Automated systems are tasked with spotting threats, deciding on a response, and acting at machine speed, which eliminates the possibility of securing human authorization for every single action.
Yossi Karadi stated that we are approaching a reality where cyber defense systems will need to make decisions at a pace that does not allow for human involvement in every action.
Autonomous Defense Systems Complicate Accountability and Decision Making
Granting greater autonomy to defense systems creates a critical paradox. Increased independence makes machine decision-making harder to interpret, more difficult to halt in time, and complicates accountability when errors occur.
The strategic document illustrates the danger through a high-stakes scenario. An autonomous cyber defense network detects a zero-day attack—an exploit targeting a vulnerability with no available security patch—and must decide whether to isolate critical infrastructure within seconds.
- A correct choice may halt the assault, but a misidentification could result in the shutdown of an essential network, such as the electricity system or a hospital.
- A correct choice successfully blocks the breach.
- A false identification risks shutting down vital networks such as hospital systems or electricity grids.
Karadi Proposes Model Balancing Innovation Security and Human Control
To navigate these operational hazards, Karadi proposes that governments adopt the AI Strategic Triangle, a model balancing innovation, security, and human control. Nations must evaluate where to accelerate AI adoption, where external vendor reliance creates national security vulnerabilities, and which authorities must never be transferred to a machine. To address these challenges, Karadi suggests that nations adopt the strategic artificial intelligence triangle, which balances three interests: innovation, security, and human control.
- Actions fully delegated to artificial intelligence systems.
- Operations permitted strictly under predefined limitations.
- Tasks requiring mandatory human authorization.
- Irreversible national decisions forbidden from autonomous delegation.
The final tier represents what the directorate terms a sovereign authority that cannot be delegated
—a sovereign authority that cannot be delegated to automated processes, keeping ultimate judgment strictly in human hands.
The Imperative for Leadership Action
Karadi emphasized that political leaders do not need to become software engineers, but they must establish firm boundaries regarding where dependence is acceptable and where national resilience is non-negotiable. They must decide where dependency is acceptable, where resilience is non-negotiable, and where human authority must remain decisive.
Yossi Karadi stated that leaders do not need to become AI engineers, they need to decide where dependency is acceptable, where resilience is non-negotiable, and where human authority must remain decisive.
The document concluded with a direct call to action, urging authorities not to leave future governance to external actors while enlisting stakeholders to construct a secure and trusted technological infrastructure. He concluded the document by urging leaders not to leave the definition of boundaries to others, calling on them not to let others define their future, but instead to enlist them in building the world they believe in.