John Ratcliffe, director of the CIA, compared artificial intelligence to “digital nuclear weapons”

Ratcliffe also presented internal changes in the intelligence agency and, despite the risk, defended a faster adoption of emerging technologies. The focus was on AI, quantum computing, biotechnology, cybersecurity and geopolitical competition with China.

“In conversations with many of the president’s other national security and economic security advisors we are talking about the impact of these frontier AI models,” Ratcliffe said, when making the nuclear comparison.

The analogy of the director of the North American intelligence center echoed in an auditorium full of Washington DCwhere AWSAmazon’s cloud computing division, hosts an AI and cloud conference for the US government ecosystem every year. And it resonated strongly because Claude Mythos Preview, Anthropic’s model with advanced cybersecurity capabilities, is offered through Amazon Bedrock, AWS’s platform for deploying generative artificial intelligence models in enterprise environments.

From the noise that Mythos Preview made (the model’s assisted – or even automatic – hacking capabilities), in fact, the system is limited to enabled organizations. Even three weeks ago, Andy Jassy, ​​CEO of Amazon, told US Treasury Secretary Scott Bessent that company researchers had managed to use an Anthropic model (Fable 5) to obtain information that could be used in cyber attacksas reported by the Wall Street Journal.

That was followed by a restriction by the US Department of Commerce on Claude Mythos 5 y Claude Fable 5which had forced Anthropic to suspend access to those models for foreign citizens. The measure, issued on June 12 for national security reasons, was lifted days later, according to The New York Times.

Ratcliffe assured that artificial intelligence, combined with other emerging technologies, is already part of the dispute for global power. “The nation that best harnesses the power of technology will determine the global future,” he said.

Their argument took direct aim at so-called frontier AI models, that is, the most advanced systems available or in development, including those underpinning commercial products such as ChatGPT, Claude or Geminibut also specialized or restricted use versions for companies, governments or technical tasks.

In his speech, Ratcliffe placed artificial intelligence within a broader transformation of war and national security. According to him, the advances in “AI, quantum computing and biotechnology” go far beyond their economic impact: “They are, as we speak, rewriting the reality of the conflict and the asymmetric war”.

To explain that change, he mentioned the use of AI-powered robots and drones. In a factory, he said, those technologies can improve speed and efficiency. On the battlefield, however, “Their efficiency and lethality can be equally devastating.”

The CIA chief linked that transformation to Ukraine, where cheap and increasingly precise drones became one of the central weapons of the conflict. “A big part of the reason for that is technology and how drones have become low-cost, super-efficient killing machines,” he said.

Now, in a darker section of his speech, the official admitted that the underlying concern is that the same advances that allow civil processes to be automated can also accelerate military operations, surveillance tasks, intelligence analysis, cyber attacks or manipulation campaigns.

Ratcliffe said America’s “adversaries” are also watching that race: “Our adversaries around the world see it too as they work to steal and manipulate America’s advances for their own ends and benefits.”

The phrase “digital nuclear weapons” did not point to the literality of an AI model having the same functioning as a nuclear bomb. What Ratcliffe pointed out was that the ability of these tools to alter balances of power is analogous to what the Manhattan Project and the arms race of the Cold War were.

For this reason, the director of the CIA insisted that whoever masters these capabilities before his rivals could gain a decisive advantage in intelligence, defense, cybersecurity and military operations.

“Each algorithmic decision has implications for the strategic advantage of the United States and for the national security of all our people,” he concluded.

The speech also functioned as a defense of the CIA’s internal transformation. In a debate crossed by technological competition with China, several sections of the keynotebeyond Ratcliffe’s intervention, insisted on the need for the United States to lead the AI ​​transformation process.

In Washington there is real concern about the development of Chinese frontier models, such as DeepSeek y Qwenand by the speed with which Beijing seeks to convert these advances into economic, military and intelligence power.

Because of this, Ratcliffe said emerging technologies and China would be the agency’s top priorities. “Eighteen months later, that approach hasn’t changed. But what has changed is the rhythm as technology evolves,” he noted.

AI adoption, he said, is already transforming the way the agency works. “Our widespread adoption of AI is literally transforming the way we do business. We’re going to take smart risks, we’re going to experiment, and then we’re going to course-correct as we go,” he said.

In turn, Ratcliffe called for moving quickly, but also drew a clear line on the importance of human intervention: final decisions must remain in human hands.

“Good intelligence will always require good judgment and only people can and should decide what is the right path to follow“, said.

The phrase is important because it separates the use of AI as an analysis, automation or operational support tool from the complete delegation of sensitive decisions. In intelligence, that limit may involve assessments of human sources, covert operations, cyber actions, military objectives, or threats against critical infrastructure.

Ratcliffe even suggested that the traditional profile of the intelligence agent is going to change. “More CIA officers will have to be as comfortable handling lines of code as they are handling human assets and sources,” he said.

The comparison with “digital nuclear weapons” thus leaves a political signal about how the CIA sees AI: as a part of the core of current geopolitical competition. The agency wants to incorporate it faster, use it in more areas and train officers capable of operating in that new environment.

But, at the same time, they have human judgment as an irreplaceable variable in the equation.