Pegasus spyware founder urges governments to trust his tech and ditch US AI
Former NSO Group boss Shalev Hulio seeks to reinvent himself as a champion of sovereign AI and national security
Countries should host their own AI models rather than entrust sensitive national data to US technology companies, according to the man once responsible for the Pegasus tool used to hack the phones of politicians, reporters and rights activists.
“Every government should have their own model,” Shalev Hulio told the FT. “They should not take the risk that somebody will wake up one day and shut their model because AI is a critical infrastructure,” he continued. “You want to control your own AI.”
The comments represent the latest pitch from the spyware salesman turned cyber security evangelist, whose new venture Dream promises to help governments safeguard their most valuable information and then harness it for the national good.
It is, perhaps, an unlikely proposition from a man who made his name in the 2010s as chief executive and co-founder of NSO Group, the infamous Israeli commercial spyware manufacturer.
Hulio has long faced fierce criticism over Pegasus, including from the then UN special rapporteur on free speech who urged him in an open letter in 2020 to take more robust action to prevent its use in some countries as a tool of surveillance and repression.
The 45-year-old expressed no regret about his previous life, seeing his years selling spyware to governments as an asset to Dream rather than a liability.
“Despite the public opinion, governments managed to prevent terror attacks and managed to save hundreds of thousands of lives along the years using NSO products,” he said. “I actually see [my past] as a huge advantage.”
Dream was established in 2022 following a chance encounter between Hulio and co-founder Sebastian Kurz, the youthful former Austrian chancellor, whose own political career had collapsed the previous year.
Initially conceived as a cyber security provider to governments, Dream was an early advocate of the need to use AI tools for cyber defence.
“American VCs thought we were cuckoo,” said Hulio of his first talks with potential investors in December 2022.
“I told them five years from now . . . the cyber attack is going to be completely autonomous and the only way to be ready is to build autonomous defenders,” he recalled. “They told me, ‘listen you’ve lost your mind’.”
Dream has since raised more than $400mn and was valued at $3bn in June. Investors include Bain Capital Ventures and Tau Capital, an Abu Dhabi-headquartered investment firm backed by Sheikh Tahnoon bin Zayed al-Nahyan, the UAE’s powerful national security adviser.
The company’s first product, Sphere, is marketed as a national-scale cyber defence system that uses AI-powered tools to protect governments and critical infrastructure against attacks by hostile states. A second product, Hero, uses AI to autonomously find and patch software vulnerabilities.
In August, Dream’s researchers found evidence of what they described as the first ever “end-to-end autonomous” attack on a government target. “We predicted that it would take five years. It took three and a half,” said Hulio.
Having originally focused on how to protect government data from cyber attacks, Hulio and Kurz decided they could also help countries use their data better.
In June, the pair launched a third product, Atlas, an AI platform designed to allow governments to develop and operate customised AI models hosted on their own infrastructure without relying on external cloud providers.
Unlike the broad capabilities of popular large language models like Anthropic’s Claude or OpenAI’s ChatGPT, Dream’s bespoke solution is trained on the government’s own data and tailored to specific tasks, like analysing classified intelligence or improving procurement. Hulio’s bet is that this tighter focus is cheaper to run and more closely aligned with what governments need.
“We are saying to entities within the government, you should have a very narrow but very smart model that can run very fast on one or two or three GPUs,” Hulio said, referring to the powerful chips used to run AI systems.
Hulio left NSO Group in 2022 after a decade at the helm. His departure followed a tumultuous few years after the FT revealed in 2019 how NSO had exploited a vulnerability in the WhatsApp messaging service to install Pegasus spyware on iPhone and Android phones.
Once valued at as much as $1.5bn, NSO was blacklisted by the US government in November 2021 and subsequently lost a landmark lawsuit brought by WhatsApp over the hacks.
Despite the controversy, Hulio says his NSO experience has left him with a unique kind of official access. “Governments really appreciate, I think, the relationship that we had before,” he said.
Headquartered in Tel Aviv, with offices in Vienna and Abu Dhabi, Dream has grown to about 350 employees in three years.
“Shalev has a fantastic black book,” said one early Dream investor, adding that Hulio’s “chutzpah” and Kurz’s government experience seemed to be “a good combination”.
Dream does not disclose which governments it works with but says it has sold its cyber security tools to more than a dozen countries, including eight in Europe, five in Asia and two in the Middle East. The sovereign AI product has been deployed in two countries so far, Hulio said.
“When you’re working with government you need a lot of trust and confidence,” Hulio added. “I think this is what we are bringing to the table before the technology . . . trust and confidence.”