CIA chief warns of AI-powered ‘digital nuclear weapons’

AI-driven cyberoffensive tools can be compared to “digital nuclear weapons,” CIA Director John Ratcliffe has said, warning that they could fuel rivalries among global powers.

Ratcliffe made the comparison on Tuesday in a speech at the Amazon Web Services summit in Washington, where he discussed the spy agency’s efforts to speed up the acquisition of private-sector products for its own use.

“AI tools will only continue to raise the stakes in our competition with all of America’s adversaries,” Ratcliffe said. It would be “not misplaced to refer to their capabilities as akin to digital nuclear weapons,” he added, citing discussions within the administration of US President Donald Trump.

Ratcliffe claimed that rival nations “work to steal and to manipulate America’s advancements for their own ends and gains.”

Promise of US AI dominance

Promises of rapid advances in AI capabilities, including in hacking, have been a constant feature of the global digital technology race. Last month, the Five Eyes intelligence alliance, comprising Australia, Canada, and New Zealand, the UK, and the US, warned that frontier models are “anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities,” adding that “the timeline is not years, it is months.”

US Sen. Mark Warner (D-VA) echoed the warning during an Intelligence Committee hearing, saying National Security Agency chief Joshua Rudd had told him that Anthropic’s Mythos 5 model “broke into almost all of our classified systems, not in weeks, but in hours.” According to the New York Times, the description “simplified” the spy agency’s controlled tests, which were aimed at AI-assisted identification of cybersecurity flaws rather than actual hacking.

The ongoing boom in the American AI sector is based on hopes of massive future profits to justify the hundreds of billions of dollars in investment. Some analysts have described it as a financial bubble, warning that it could crash unless US tech giants achieve global dominance.

Not all predictions about what AI can do have materialized. While current models are highly competitive in computer coding and data analysis, for instance, fully autonomous driving remains years behind the timeframes Tesla CEO Elon Musk presented to the public in the past.

‘DeepSeek moments’ threaten American goals

A major risk to the US AI push is foreign competition able to deliver products comparable in power at far greater efficiency.

China’s DeepSeek app sent shockwaves through the industry in January 2025, when its R1 and V3 models proved comparable to contemporary digital engines used by ChatGPT and other US rivals, but at a fraction of the cost. US officials claimed the Chinese company had essentially cheated by building its product based on American work.

A similar DeepSeek moment followed the release of Zhipu’s new flagship coding assistant model, GLM-5.2, in mid-June, the South China Morning Post reported last week. Matt Velloso, a former vice president at Meta Platforms and Google DeepMind, described it as the “first open model that passes the bar as a daily driver.”

China is quickly catching up with the US in manufacturing advanced microchips, which limits Washington’s ability to use trade restrictions to slow down its competitors. Meanwhile, China’s relative abundance of energy generation gives it an advantage in the AI race.