SI Agent News
DevelopingPolicySecurity

The Pegasus spyware co-founder now tells governments to build their own AI

Shalev Hulio, who co-founded NSO Group, argues countries should run their own AI models instead of trusting US firms, and he now sells the tools to do it.

$3 billionDream's valuation earlier this year

The short version

Why it matters

Governments are being pushed to treat AI like power grids or telecoms: something you want to control, not rent. The messenger has a history, and a product to sell, so the argument deserves scrutiny. But the underlying worry about depending on a foreign supplier is one many officials share.

A surprising voice on AI independence

As TechRadar on MSN reported, Shalev Hulio is urging governments to stop leaning on American AI companies for sensitive work. Hulio is the former chief executive and co-founder of NSO Group, the Israeli company behind Pegasus, a piece of spyware (software that secretly monitors a phone or computer). Pegasus was used to target politicians, business leaders and journalists around the world.

Hulio now runs a different kind of business. In 2022 he co-founded Dream, an AI cyber defence company, with former Austrian chancellor Sebastian Kurz. His message to governments is that national security data should not sit with outside AI providers.

He put it bluntly to the Financial Times: "Every government should have their own model."

The case for owning your AI

Hulio's core argument is about dependence. In his telling, AI has become critical infrastructure, meaning something a country needs to keep working, much like electricity. If a government rents that capability from a foreign company, it takes on the risk that the company could one day cut off access to its model.

This is what people mean by sovereign AI: a country running AI systems it controls, on infrastructure it controls. It is our shorthand, not the article's, but in practice the idea comes down to governments wanting models they can host and operate themselves.

The pitch is not about building the biggest chatbot. Hulio says government bodies need models that are narrow in scope but smart in that scope, and fast enough to run on one, two or three GPUs. A GPU is a type of chip that does the heavy math AI needs, so that is a small hardware footprint compared with giant commercial systems.

What Dream sells

Dream began by offering AI-enabled cyber defence tools to governments. It now lists three kinds of products:

The models are made to order for government use and trained on each customer's own data. According to the report, that tight fit to government needs also helps keep costs down.

The company has raised $400 million in funding and investment, and it was valued at $3 billion earlier this year. It says it has sold to over a dozen countries across Europe, Asia and the Middle East. Dream has also been credited with spotting what is described as the first autonomous, end-to-end AI attack. In that case, Chinese hackers used open-source AI models to find and chain together software flaws and break into Taiwanese government agencies, including the nuclear safety agency.

The baggage behind the message

Hulio's history is hard to separate from his advice. The US government blacklisted NSO Group, and the UN and the EU have continued to criticise it over Pegasus being put on thousands of devices through a flaw in WhatsApp.

Hulio defends that record. He told the Financial Times that governments used NSO products to prevent terror attacks and save hundreds of thousands of lives over the years, and that governments valued the relationship they had with his old company. Those are his claims; the article does not offer evidence to check them, and critics dispute the wider picture.

There is also a plain commercial point. Hulio is telling governments to buy something his company happens to sell. That does not make the warning wrong, but readers should weigh it as a sales pitch as well as a policy view.

What to take from it

The worry underneath is real and not unique to Hulio: if your most sensitive work runs on someone else's AI, your options narrow when that supplier changes terms, prices or access. The same logic applies to a small business that builds everything on one vendor, only with lower stakes.

What is still unclear is how well small, custom government models perform against the large commercial ones, and what it costs to maintain them over time. The source gives no performance figures or customer names, so those claims remain Dream's own.

What to watch

Try it

List the AI services your team depends on, and write down what you would do if one of them cut off access tomorrow.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by TechRadar on MSN · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news