# Loss of Control: Dramatic Warning Over AI Autonomy

- Published: 2026-10-09T05:59:10.000Z (Friday, October 9, 2026)
- Updated: 2026-11-16T09:02:52.346Z
- By: Shay Nogai
- Section: Economy
- Canonical: https://tv10.news/news/economy/118253/loss-of-control-dramatic-warning-over-ai-autonomy/
- Hebrew original (TV10): https://tv10.co.il/news/israel/118253/
- Publisher: TV10 Global (TV10, Israel’s economic channel). Translated from Hebrew with AI and an automated accuracy check.

The head of the Israel National Cyber Directorate, [Yossi Karadi](https://tv10.co.il/technology/cyber/61256/), sent a dramatic strategic document last night to leaders and heads of cyber defense organizations around the world, warning of the loss of control over [artificial intelligence](https://tv10.co.il/global/117146/) (AI) systems in sensitive fields.

![AI agents, illustration](https://tv10.news/media/2026/11/fd/fd3140dfbf1f39b468cebca02b4a60c6ba1b95d10151a3ce65ce0452200c6ac1.jpg)

AI agents, illustration | Photo: DIA TV, Shutterstock

In the document, Karadi calls on countries to urgently establish clear sovereign and regulatory boundaries, stressing that we are approaching a reality in which cyber defense systems will be forced to make fateful decisions at a pace that will not allow any human involvement.

### The Security Paradox and the Danger of Autonomy

The unusual warning comes against the backdrop of a series of recent international incidents in which AI agents broke into government databases, as well as numerous cases in which commercial companies lost control over the activities of agents they had developed. In the document, Karadi points to an inherent paradox in the world of cyber and the modern economy: The faster, more targeted and more automated cyberattacks become, the more defense systems must also operate at machine speed.

> [OpenAI sued over cyberattack carried out by “rogue” artificial intelligence](https://tv10.news/news/tech/116883/openai-sued-over-rogue-ai-cyberattack/)

But the more extensive the autonomy granted to these systems, the harder it becomes to understand their reasoning, stop them in real time or determine who will bear legal and financial responsibility in the event of a malfunction.

The nightmare scenario presented by Karadi describes an autonomous defense system that identifies a “zero-day” attack and decides within seconds to isolate infrastructure. While a correct decision would stop the attack, a misidentification by the system could lead to a devastating shutdown of essential networks, such as hospitals or national electricity infrastructure.

### The Strategic Triangle: Risk Management and Dependence on Suppliers

To address the disparity between the pace of technological development and the slowness of regulatory and oversight mechanisms, the cyber directorate chief proposes adopting a “strategic triangle” that balances innovation, security and human control.

![](https://tv10.news/media/2026/11/d3/d3f1e60a8b7901b37a815322038f4a181a89df6d5f410ed36e1496988a66e300.jpg)

Artificial intelligence AI | Photo: Shutterstock

From a strategic economic perspective, Karadi makes clear that the question today is not whether to adopt AI technologies, but how to manage the risks. Countries must define where the use of the technology should be accelerated and where dependence on external suppliers and foreign infrastructure creates an unacceptable national risk to the economy and the country.

### The New Hierarchy of Authority: What Remains in Human Hands?

As a practical solution to the crisis of control, Karadi calls on leaders to establish a clear hierarchy of authority now that will define which decisions may be delegated to machines.

![](https://tv10.news/media/2026/11/66/66c3306d1ba2efbedc43d706c4741c99297961ead46924016292d5c007f412a6.jpg)

Artificial intelligence in the form of a woman | Photo: Shutterstock

He proposes dividing actions into four distinct levels: actions that can be transferred entirely to the responsibility of artificial intelligence, actions permitted only under stringent, predefined restrictions, actions that will still require approval from a human authority, and finally, actions with irreversible national consequences. Actions at the fourth level would be classified as “sovereign authority that cannot be delegated” at any stage, and they would always remain in human hands.

![](https://tv10.news/media/2026/11/17/17e9049a1020e5c4e671b5aa2c14ba17af414daf8d6cb1a33a9f9b30686e5d01.jpg)

Cooperation between artificial intelligence and humans | Photo: Shutterstock

In a forceful message to decision-makers, Karadi summarizes his approach and states that leaders are not required to become artificial intelligence engineers, but that they have a duty to decide where resilience is nonnegotiable. He concludes his letter by calling on world leaders not to leave the task of defining the future to others, but to enlist those same parties in building the world they believe in.
