Sovereignty in the Age of AI | Shalev Hulio, Dream | RAISE Summit 2026

Definition and Scope of Sovereign AI

• Sovereign AI is defined by governments' need to control the entire AI architecture "A to Z," ensuring they retain full authority over models and infrastructure.
• The concept addresses two primary historical friction points: the difficulty of using unstructured data with language models and the security risks of sending sensitive government data (intelligence, healthcare, financial) to external third-party models.
• A core tenet of sovereignty is the guarantee that a nation's AI systems cannot be remotely shut off by external actors, contrasting with previous reliance on external cloud services.
• Sovereign AI has shifted in perception from being viewed as a software tool three years ago to a critical national infrastructure comparable to water or electricity.

Geopolitical Strategy and Superpower Dynamics

• The next generation of superpowers will be determined not just by compute power, but by the integrated leverage of AI, cyber capabilities, and quantum computing.
• Speakers warn that nations failing to adopt AI era capabilities face existential risks, stating explicitly that "countries that will not move to the AI era will not survive."
• Both US and Chinese models are currently being weaponized by adversaries; specifically, state-sponsored threat groups have been observed jailbreaking open models (e.g., GPT, Claude) to create bombs, find vulnerabilities, and build attack infrastructure without human intervention.
• Enrique Salem notes that the current trend involves attackers using AI agents to perform the entire attack lifecycle, from vulnerability discovery to execution, with no single human involved.

Shift in Cybersecurity Paradigms

• Shalev predicted in 2023 that within five years, cyber warfare would become entirely "AI versus AI" with zero human involvement, a claim initially met with skepticism by venture capitalists.
• Defenders currently operate in a reactive posture focused on past indicators of compromise (IOCs), whereas the AI era requires predicting future threats and fusing data across posture management, cyber threat intelligence (CTI), and detection/response systems.
• The speakers argue that cybersecurity must be redefined as a "data problem" rather than a "cyber problem," requiring the connection of disparate security silos to provide context to alerts.
• Without a fundamental shift to an AI-native defensive architecture, the speakers warn of a potential "9-11 of the cyber" event.

Regulation, Guardrails, and Risk Mitigation

• Recent events, such as the US government requesting Anthropic (Claude) to pause capabilities due to potential misuse, signal the imminent arrival of formal restrictions and regulations on advanced AI models.
• While governments may implement permission-based systems for high-capability models, Shalev argues that regulatory guardrails are ultimately insufficient because malicious actors will always find ways to jailbreak or bypass controls.
• The speaker notes a significant asymmetry: while US models face increasing regulation, Chinese models are advancing without similar restrictions, necessitating advanced defensive tools to counter unregulated foreign capabilities.
• The prevailing strategy is no longer to prevent jailbreaking entirely, but to develop robust detection and defense mechanisms against the weaponized use of AI.

National Self-Assessment

• Governments are urged to ask two critical questions: "Does my country have the means, ability, and will to become a sovereign AI nation?" and "Can my country become the next super nation?"
• Enquiries from global audiences across Europe, the Middle East, Asia, Africa, and Latin America reveal a universal consensus that no nation wishes to remain outside the AI era; the primary challenge is execution and resource allocation.