Why is the push for sovereign AI becoming a national security priority?
The movement toward sovereign AI is driven by the need to protect sensitive government data from being processed by foreign entities. Shalev Hulio, the former chief executive and co-founder of the Israeli spyware firm NSO Group, has stated that governments should not trust US AI companies with their most sensitive datasets. The core argument is that AI has transitioned from a mere productivity tool to a fundamental component of critical national infrastructure.
Hulio's warning to the Financial Times emphasizes the geopolitical risk of dependency. He suggests that relying on external providers creates a vulnerability where a foreign power or corporation could unilaterally decide to deactivate access to essential models. By developing and hosting their own models, nations can ensure continuity of service and maintain absolute control over the logic and data driving their state functions.
The risk of external model termination
The concept of 'model shutoff' represents a significant strategic threat. If a nation's administrative, military, or intelligence functions are built upon proprietary models owned by a foreign corporation, that nation effectively cedes a degree of its sovereignty. In a period of geopolitical tension, the ability of an external provider to 'wake up one day and shut their model' could paralyze a government's digital capabilities.
How does Dream aim to facilitate national AI independence?
Dream, an AI cyber defense firm co-founded by Hulio and former Austrian chancellor Sebastian Kurz in 2022, provides the technical framework for this transition. The company offers a specialized portfolio designed to help governments build bespoke AI models that operate on domestic infrastructure. This approach is intended to move away from the 'one-size-fits-all' models provided by major US tech giants, which often require massive, centralized computing power.
Instead of massive, general-purpose models, Dream advocates for highly specialized, efficient systems. Hulio notes that government entities should utilize 'very narrow but very smart' models. These specialized models are designed to run efficiently on a minimal number of GPUs—specifically one, two, or three—making them more accessible for sovereign deployment within controlled, local environments. This efficiency helps reduce the total cost of ownership while increasing the speed of response for critical tasks.
Bespoke training and cost efficiency
A key component of the Dream strategy is the use of customer-specific data to refine these models. By training AI on a government's own data, the resulting tools become more tightly aligned with specific national needs and regulatory requirements. This bespoke training process not only improves accuracy in specialized fields like intelligence or defense but also helps keep operational costs down by avoiding the overhead associated with massive, unoptimized general models.
What role does AI play in modern cyber defense?
AI has become a central actor in the landscape of digital warfare, capable of both launching and defending against sophisticated attacks. Dream has already gained recognition for its role in detecting high-level threats, including the world's first autonomous AI end-to-end attack. In that instance, Chinese hackers utilized open-source AI models to identify and exploit a chain of vulnerabilities to penetrate Taiwanese government agencies, including a nuclear safety agency.
To counter such threats, Dream offers several AI-driven layers of protection:
- National-scale cyber defense systems: AI-enhanced frameworks designed to monitor and protect an entire nation's digital perimeter.
- Software vulnerability discovery: An AI-powered platform that proactively searches for weaknesses in code before they can be exploited by adversaries.
- Sovereign AI deployment: Tools that allow governments to set up their own AI ecosystems without relying on external, potentially compromised, infrastructure.
What is the controversy surrounding Shalev Hulio and NSO Group?
The advocacy for sovereign AI by Hulio comes against a backdrop of significant international scrutiny regarding his previous work. As the head of NSO Group, Hulio was responsible for the development of Pegasus, a commercial spyware capable of monitoring the communications of politicians, journalists, and business leaders globally. Pegasus gained notoriety for exploiting flaws in applications like WhatsApp to gain access to thousands of devices.
This history has led to various forms of institutional pushback. The NSO Group has faced blacklisting by the US government and has been the subject of ongoing criticism from both the United Nations and the European Union. Despite these criticisms, Hulio maintains a different perspective on the utility of such technology. He told the Financial Times that, notwithstanding public opinion, governments have successfully used NSO products to prevent terrorist attacks and save hundreds of thousands of lives over the years.
The tension between surveillance and security
The debate surrounding Hulio’s career highlights a fundamental tension in modern security: the line between legitimate state intelligence gathering and the infringement of privacy. While Hulio argues that the relationship between spyware providers and governments is valued for its life-saving capabilities, international bodies continue to raise alarms about the potential for these tools to be used for political repression and the targeting of civil society.
How much investment is flowing into sovereign AI defense?
The financial backing for companies like Dream suggests that the market for sovereign AI and cyber defense is expanding rapidly. Dream has successfully raised $400 million in funding and investment, reaching a valuation of $3 billion earlier this year. This level of capital indicates strong investor confidence in the necessity of domestic AI infrastructure for national security purposes.
The commercial footprint of these technologies is also growing. Dream has already sold its suite of products to more than a dozen countries across Europe, Asia, and the Middle East. This rapid adoption suggests that many governments are already moving toward the model of localized, AI-driven defense to mitigate the risks posed by both traditional cyber threats and the emerging era of AI-enabled warfare.
Frequently asked questions
What is the difference between general AI and sovereign AI?
General AI refers to large-scale models provided by external corporations, often hosted on foreign servers. Sovereign AI involves a nation developing and hosting its own models on domestic infrastructure, ensuring that the data, the training processes, and the control of the model remain entirely within national jurisdiction.
Why can't governments just use US-based AI services?
Using US-based services introduces risks of data exposure and service dependency. If a government relies on external AI for critical functions, it risks losing access if the provider changes its policies or if geopolitical tensions lead to the sudden termination of service, potentially paralyzing national infrastructure.
How does Dream make AI more efficient for governments?
Dream focuses on creating 'narrow but smart' models. Instead of massive, resource-heavy systems, these models are optimized to perform specific tasks highly effectively while requiring very little hardware, such as only a few GPUs, making them easier to host and manage locally.
Has AI been used in actual cyber attacks?
Yes, AI has been used in autonomous attacks. For example, Chinese hackers reportedly used open-source AI models to discover and chain vulnerabilities to breach Taiwanese government agencies, including a nuclear safety agency, marking a significant escalation in the use of AI in cyber warfare.
What was the significance of Pegasus spyware?
Pegasus was a commercial spyware developed by NSO Group that could exploit vulnerabilities in mobile devices to monitor users. It became highly controversial due to its use in targeting journalists, politicians, and activists, leading to sanctions from the US and criticism from the UN and EU.
Key takeaways
- Sovereign AI is essential for preventing foreign entities from controlling a nation's critical digital infrastructure.
- Shalev Hulio argues that governments must host their own AI models to avoid the risk of sudden service termination.
- Dream provides specialized, narrow AI models that can run on minimal GPU hardware for efficient local deployment.
- AI-enabled attacks, such as those seen in Taiwan, demonstrate the urgent need for autonomous cyber defense.
- The market for sovereign AI defense is growing, with Dream valued at $3 billion and serving over a dozen countries.
The future of national digital autonomy
The shift toward sovereign AI represents a fundamental change in how nations perceive digital sovereignty. As artificial intelligence becomes deeply integrated into the fabric of statecraft and national defense, the ability to control one's own computational intelligence is no longer a luxury, but a necessity. The transition from general-purpose, foreign-hosted models to specialized, domestically-controlled systems will likely define the next era of international relations and cybersecurity. For governments, the choice is between the convenience of external providers and the security of true technological autonomy.
